Skip to main content
About Me

Professional Experience

TIH Derivco DVT Seecrypt Covariant Consulting UD Trucks Enermatics Infotech Information Trade World City of Tshwane Marnet Signs Books-u-profit Nexor Signs

Education

Diploma in C++ Programming Grade 12 (Matric)

Technical Skills

Languages Frameworks & Methodologies Architectural Concepts

Projects

Open Source Featured Projects Personal Interests References Download CV

v1.8.1, updated 10 Oct 2026

A photo of Logan Venter Download CV

Chat with an AI agent that knows about my experience, skills, and projects.

Try It

About Me

I architect and build production AI platforms. Over 20 years in software development, from multi-protocol server frameworks
  • 5 transports: stdio, REST, Streamable HTTP, SignalR, gRPC
  • Session management with JWT and API key authentication
  • Automatic tool discovery and JSON schema validation
View project →
to agentic chatbot platforms
  • LangChain/LangGraph orchestration with pluggable LLM providers
  • Planning coordination: ReAct, Plan-Execute, Adaptive strategy selection
  • Multi-turn security pipeline with prompt injection defense
  • React and TypeScript frontends, including an embeddable chat widget
View project →
to RAG-powered knowledge bases
  • 14 chunking strategies for different document types
  • Hybrid retrieval combining BM25 keyword matching with dense vector search
  • RAPTOR hierarchical summarization for deep context
  • Trustworthy generation with hallucination detection
View project →
. My current focus: enabling organizations to deploy LLM-powered systems
  • Multi-provider routing with factory-based gateway selection
  • Structured output generation with retry and validation
  • OpenTelemetry observability with correlation IDs
  • A/B testing frameworks for model evaluation
View experience →
with enterprise-grade reliability.

What I've Built

  • tawk
    • Native terminal client in C and ncurses, linked to your phone like WhatsApp Web
    • Two interchangeable WhatsApp backends behind one contract
    • Up to eight accounts, an encrypted local database, voice note transcripts, media and statuses
    • Agent rules per chat, TL;DR summaries, and an owner’s chat for instructing your agent from your phone
    • github.com/loganventer/tawk
    C ncurses SQLite Open Source
    View project →
    , an open source WhatsApp client for the terminal, and tawk-mcp
    • Lets AI clients read chats through tawk and act with your approval
    • Destructive actions need two confirmations, one of which the model can never give
    • More than 90 tools, including transcripts, summaries, labels and reminders
    • Six iDesign layers on .NET 10 and the official MCP C# SDK
    • github.com/loganventer/tawk-mcp
    .NET 10 C# MCP Open Source
    View project →
    , its MCP server with human approval built in
  • Agentic AI framework
    • 18 independently versioned Python packages and 17 .NET projects
    • A gRPC sidecar exposes the Python reference implementation to .NET through typed clients
    • Agent tool loop, planning, RAG, security, auth, and MCP and A2A interop
    • Coverage gates per package, at 95% for most
    Python .NET 10 gRPC MCP A2A
    View project →
    for Python and .NET, covering the tool loop, planning, RAG, security and agent interop
  • Multi-agent systems
    • Code and documentation review with up to 37 parallel reviewers
    • An agent that builds agents: a twelve-phase pipeline with 838 tests at 98% coverage
    • A planner and proof gate that re-runs each step's acceptance checks in its own sandbox
    • A use-case triage and SDLC document pipeline that cites every statement
    Python A2A MCP iDesign
    View projects →
    for code review, agent generation, planning with proof, and document pipelines
  • A discontinued MCP server framework
    • 5 transports: stdio, REST, Streamable HTTP, SignalR, gRPC
    • Discontinued: the official C# MCP SDK, maintained in collaboration with Microsoft, replaced it, and the servers were migrated to .NET 10
    • JWT and API key authentication with session management
    • JSON schema validation for tool inputs and automatic tool discovery from assemblies
    • LangChain/LangGraph integration with Python client support
    .NET 9 C# gRPC SignalR MCP
    View project →
    for .NET with multi-transport support, session management, and automatic tool discovery, built before an official SDK matured
  • Agentic chatbot platform
    • LangChain/LangGraph orchestration with pluggable LLM and embedding providers
    • Production RAG with 14 chunking strategies, hybrid retrieval, and hallucination detection
    • Multi-turn security pipeline with prompt injection defense and risk-based capability restriction
    • Streaming responses via SSE and WebSocket with real-time audio transcription
    • Self-service admin portal for bot lifecycle management and knowledge versioning
    • React and TypeScript frontends, including an embeddable chat widget
    Python LangChain FastAPI React TypeScript
    View project →
    with LLM orchestration, production RAG, multi-turn security, and React frontends
  • MCP meta-tooling: servers that scaffold, build, and publish new MCP server projects from templates
  • Knowledge base platforms
    • Hierarchical document organization with tagging and audit logging
    • Semantic search via vector embeddings with dynamic Qdrant container management
    • Conversational RAG interface for natural language queries
    • Admin dashboard with file upload and version management
    .NET 9 Python Qdrant React
    View project →
    with semantic search, hierarchical document management, and conversational RAG interfaces
  • MCP servers for Azure DevOps (74 tools), Microsoft Teams, Microsoft To Do, Databricks, Toggl Track and web research
  • A governed AI gateway on .NET 10 with per-application permissions, budgets, metering and audit

Technical Foundation

Languages: C# (.NET 10), Python, C, TypeScript/React, Dart/Flutter
  • C# / .NET 10 — Primary. High-performance services, MCP servers and frameworks, AI gateways, enterprise backend systems
  • Python — Primary. LangChain, LangGraph, FastAPI, Pydantic, AI/ML pipelines
  • TypeScript / React — Strong. Modern frontends, real-time chat interfaces, admin dashboards
  • Dart / Flutter — Strong. Cross-platform mobile and desktop applications
  • SQL — Strong. T-SQL, complex schemas, query optimization
  • C — Strong. Native terminal applications with ncurses, SQLite and multi-threaded networking
  • C++ — Experienced. Performance-critical and security applications
  • Legacy — COBOL, Delphi, Visual Basic
View all languages →
Architecture: iDesign Methodology
  • iDesign methodology with strict layered dependency direction
  • Protocol-based dependency inversion and composition over inheritance
  • Domain-Driven Design with bounded contexts
  • Strategy, Factory, Pipeline, Registry, and Adapter patterns
  • Microservices and event-driven architecture
View all concepts →
(trained by Juval Löwy), Clean Architecture, DDD
AI/ML: LangChain, LangGraph, RAG pipelines
  • LLM orchestration with multi-provider routing and streaming
  • RAG: 14 chunking strategies, hybrid retrieval, RAPTOR, RRF fusion, HyDE
  • Agentic patterns: ReAct, Plan-Execute, Adaptive planning, tool calling
  • Security: prompt injection defense, canary tokens, risk scoring, spotlighting
  • Vector databases: FAISS, Qdrant, Pinecone, Chroma, Weaviate
  • Multi-agent orchestration over MCP and A2A, with proof-gated verification
View all frameworks →
, multi-agent orchestration, prompt engineering
Infrastructure: Docker, Azure, multi-protocol APIs
  • Docker and Docker Compose for containerized deployments
  • Azure cloud infrastructure and Azure AD authentication
  • Multi-protocol backends: gRPC, SignalR, SSE, Streamable HTTP, WebSocket, REST
  • OpenTelemetry + Elasticsearch observability with structured logging
  • CI/CD pipelines, cross-platform deployment (Linux, Mac, Windows)
View all frameworks →
, OpenTelemetry

What I Believe In

  • Clean architecture and deliberate design decisions
  • Deterministic systems with well-defined failure boundaries
  • Code that communicates intent
  • Direct feedback and honest technical discourse
logan.venter@outlook.com
+27 60 974 2113 / +27 76 414 0083

Senior AI/Platform Engineer

Telesure Investment Holdings (TIH) | Aug 2025 - Present Permanent

  1. Contractor Aug 2025 to Apr 2026 Placed by DotCom Software Solutions
  2. Permanent 1 May 2026 to Present Employed directly by Telesure
  • Architected a multi-project agentic AI platform
    • Layered architecture: 18 independently versioned Python packages and 17 .NET projects, with a chatbot service and administration portal built on top
    • A gRPC sidecar exposes the Python reference implementation to .NET through typed clients
    • Protocol interfaces for all major contracts — gateways, engines, accessors, registries, routers
    • Topological dependency resolution with circular dependency validation
    • Strategy, Factory, Pipeline, Registry, and Adapter patterns throughout
    iDesign Clean Architecture DDD SOLID Dependency Inversion
    applying iDesign methodology with strict dependency direction and protocol-based dependency inversion
  • Designed and built production RAG pipelines
    • 14 chunking strategies: recursive, semantic, sentence, fixed, markdown, HTML, code, table-aware, hierarchical, contextual, proposition, late, agentic, compression
    • Hybrid retrieval combining dense vector search with BM25 keyword matching
    • Score fusion via Reciprocal Rank Fusion (RRF) and weighted strategies
    • RAPTOR hierarchical summarization for deep context understanding
    • Trustworthy generation with confidence scoring and hallucination detection
    • Fluent builder API with presets (Conversational, Production, High-Accuracy, Code Search, Research)
    • Knowledge versioning with incremental updates and content hashing for change detection
    RAG BM25 RRF RAPTOR HyDE Vector Search
    with hybrid retrieval, hierarchical summarization, and hallucination detection
  • Implemented agentic planning and orchestration
    • Planning coordinator with 3 strategies: ReAct (iterative reasoning), Plan-Execute (upfront planning), Adaptive (LLM auto-selects based on task complexity)
    • Tool registry with automatic discovery, execution with retry logic, and canary token support
    • Multi-model routing by engine type (agentic, planning, reflection, classification)
    • MCP tool integration for external tool server composition
    • Vector-backed conversation memory with semantic similarity
    • Rolling window context management with fallback summarization
    ReAct Plan-Execute Tool Calling MCP LangChain LangGraph
    with tool calling, multi-model routing, and conversation memory
  • Applied enterprise security patterns
    • Pluggable security pipeline with scanner adapter registry
    • Multi-turn attack detection across conversation history
    • Prompt injection scanning with pattern-based and semantic detection
    • Risk accumulation and decay model with configurable thresholds
    • Canary token management for data exfiltration detection
    • Capability restriction and human escalation at risk thresholds
    • Tool chain validation to prevent unauthorized operation sequences
    Prompt Injection Defense Canary Tokens Risk Scoring Spotlighting
    : multi-turn attack detection, prompt injection defense, and risk-based capability restriction
  • Integrated multiple LLM and embedding providers
    • 6 LLM gateway implementations with unified interface: streaming and non-streaming generation
    • 3 embedding gateway implementations with unified embed() contract
    • Factory pattern for runtime provider selection from configuration
    • 7 vector store integrations with semantic caching layer
    Anthropic Azure OpenAI Ollama FAISS Qdrant Pinecone Chroma Weaviate
    with factory-based routing and graceful degradation
  • Built a self-service administration portal
    • Multi-step bot creation wizard with real-time validation and SSE progress streaming
    • LLM-powered auto-generation of bot personality, system prompt, and greetings from document analysis
    • Knowledge base versioning with incremental updates: delta detection, embedding reuse for unchanged content
    • Agent deployment management with port allocation and health monitoring
    • In-browser chat playground for testing deployed agents
    • Dashboard with health status, usage metrics, and error logs
    React TypeScript FastAPI SSE Vite
    for bot lifecycle management, knowledge versioning, and deployment orchestration
  • Designed observability infrastructure
    • Multi-destination logging: console, file, and Elasticsearch via OpenTelemetry
    • Request correlation IDs propagated across service boundaries
    • A/B testing framework with experiment/variant management and assignment tracking
    • Centralized Elasticsearch + Kibana stack shared across all services
    OpenTelemetry Elasticsearch Kibana A/B Testing
    with structured logging, correlation IDs, and A/B testing metrics
  • Built production chatbot service
    • SSE and WebSocket streaming for real-time response delivery
    • Real-time audio transcription via WebSocket
    • Multi-bot routing with hot-reload YAML-based bot discovery
    • Side-by-side A/B response comparison for model evaluation
    • Conversation persistence with SQLite and rolling window context management
    • Multi-provider auth middleware (Azure AD, JWT, API key) with RBAC
    FastAPI SSE WebSocket React RBAC Docker
    with streaming responses, real-time audio transcription, multi-bot routing, and an embeddable React chat widget
  • Built multi-agent systems on the A2A protocol: code and documentation review agents
    • Three review intensities, running 9, 27 or 37 specialised reviewers in parallel over a pull request
    • 77 agent specifications covering 8 backend languages, 4 frontend frameworks and the OWASP Top 10 CI/CD Security Risks
    • Six-step pipeline: trigger, review, decide, verify, deduplicate, publish, with validators iterating to convergence
    • Findings corroborated by static analysis and reachability into a composite confidence score
    • A sibling system reviews documentation against the source code it describes, across four dimensions
    • Model profiles (fast, standard, deep) that decouple agent specifications from vendor models
    Python C# MCP A2A BM25 RAG
    , an agent that builds agents
    • Turns a natural-language specification into a standalone, containerised Python agent project
    • A typed twelve-phase pipeline with 27 MCP tools, 4 guard layers and 6 archetype tiers
    • The model supplies typed structured output and the server renders the source, so it never free-writes code that has to compile
    • Phase executors verify completion claims against validators and make retries idempotent
    • 838 tests at 98% coverage
    Python FastAPI .NET MCP JSON Schema
    , a planner and proof gate
    • A calling agent asks for a plan, does a step, and submits its work as a patch
    • The service rebuilds the workspace in its own sandbox and re-runs the acceptance checks, including hidden ones
    • No API call lets the caller mark its own work done, an invariant asserted by a test
    • Ten agents: a supervisor with a state machine, a planning orchestrator running four lenses, a plan reviewer, a proof adjudicator and a blind completion auditor
    Python A2A .NET 10 MCP C# SDK
    , and a use-case triage and SDLC document pipeline
    • 19 triage agents assess a proposed AI use case for data sensitivity, reuse and build effort, and return one grounded verdict
    • Orchestration is deterministic code, and LLM judgement lives only in the specialists and the synthesiser
    • A golden-set regression suite of ideas with known verdicts, and refusal on low confidence
    • 15 document agents turn an approved verdict into six SDLC documents, each mapping every statement to its source
    • Word documents rendered by filling an existing template through Open XML
    Python A2A MCP FAISS Open XML
  • Built a terminal AI coding assistant
    • Full-screen terminal interface with 10 built-in tools for files, shell, search, git and the web
    • Permission gate with allow, deny, always and never decisions and a diff preview
    • Persistent project memory, planning, and session save, resume and rewind
    • Registers MCP tool servers and delegates to other agents over A2A, as both client and server
    • Runs headless behind a VS Code extension, with a composition test for each entry point
    Python prompt_toolkit MCP A2A VS Code extension
    with permission-gated tools, persistent memory, and agent-to-agent delegation
  • Built a suite of MCP servers
    • Azure DevOps: 74 typed tools, rebuilt on .NET 10 and the official MCP C# SDK with an 80% coverage gate
    • Microsoft Teams: about 70 tools through Microsoft Graph, in six layers enforced at compile time
    • Microsoft To Do: delegated sign-in and name-based list resolution, with 116 offline tests
    • Databricks: roughly 171 tools across 13 areas, split into read, write and admin servers
    • Toggl Track: 79 tools covering timers, projects, tasks, reports and insights
    • Web research: several search providers merged by reciprocal rank fusion behind an SSRF guard and a site allow-list
    .NET 10 C# MCP C# SDK Microsoft Graph
    giving AI agents typed, authenticated access to DevOps, collaboration, data and research platforms, with human approval for writes
    • Approval runs through MCP elicitation, so the model cannot approve its own actions
    • Destructive writes take two confirmations
    • Three access levels: read by default, then send, then manage
    • Third-party text is fenced between marker lines and labelled as untrusted
    Human-in-the-loop Prompt Injection Defense
    and shared Microsoft Entra sign-in libraries
    • Three layered .NET libraries: core authentication, Microsoft Graph, and MCP sign-in tools
    • Ten credential strategies, with the catch-all default credential deliberately left out in favour of an explicit chain
    • A token refresh service that coalesces concurrent renewals into one and serves the cached token through transient failures
    • Start-up validation of every setting, and no silent plaintext token cache
    • An 80% coverage gate with warnings as errors
    .NET Azure.Identity Microsoft Graph MCP
  • Authored an MCP server framework
    • Built before an official SDK matured: 5 transports (stdio, REST, Streamable HTTP, SignalR, gRPC)
    • Attribute-based tool registration, JSON Schema validation, and per-request API key and JWT authentication
    • Rate limiting with 3 strategies, a circuit breaker and caching
    • Adopted by other engineers to build servers over line-of-business systems
    • Now discontinued: the official C# MCP SDK replaced it, with a documented migration path
    .NET 9 C# gRPC SignalR Polly
    for .NET, then led its retirement in favour of the official C# MCP SDK and migrated the servers to .NET 10
  • Designed and built a governed AI gateway
    • Chat, embeddings, images, speech, document reading, OCR and retrieval behind one API
    • Identity, role-based permissions, per-application budgets, usage metering and a request audit
    • A governance pipeline that checks permission and then budget before any model call
    • Six iDesign layers enforced by the compiler, with a stored-procedure-only data layer guarded by a convention test
    • Exactly-once settlement of streamed responses, with one code path for streamed and buffered calls
    • Failover across named model connections, and a job queue with a background worker for deferred work
    .NET 10 ASP.NET Core SQL Server OpenTelemetry
    on .NET 10 with per-application permissions, budgets, metering and audit
  • Architected and designed an agentic knowledge ingestion and curation pipeline
    • An orchestrator runs every mechanical stage in process and calls four single-purpose agents over A2A
    • Agents are definitions and not code: a card, a system prompt and a template in a folder
    • Reviewable text and the built index go to different places, tied together by one committed pointer
    • A fail-closed sensitive-data gate runs before indexing, and index files are verified before they are opened
    • A new assistant is purely a configuration file, with no per-assistant code
    Python A2A RAG Config as Code
    that keeps assistant knowledge bases current, with human review through pull requests
  • Ran an automated CI/CD control review
    • 49 projects and 1,081 repositories reviewed against 17 fixed checks
    • Build definitions, pipeline YAML, branch policies and build timelines read through the REST API
    • Shared pipeline templates resolved across repository boundaries, with conditions evaluated
    • Every finding attributed to its origin: a shared template, a repository override, or the repository itself
    • Per-repository reports, project and estate roll-ups, and a steering committee briefing
    Azure DevOps REST API YAML pipelines DevSecOps
    across more than 1,000 repositories, attributing every finding to its origin
  • Wrote the developer onboarding course
    • An eight-chapter guide to the framework, with a runnable playground
    • Two worked examples and six exercises: security, memory, MCP tools, planning, streaming, and a composed agent
    • Each exercise is a stub plus the tests that define done
    • One CI step per exercise, so the build reads as a checklist, with an 80% coverage floor
    Python pytest CI
    for the framework, with test-graded exercises
  • Leading technical architecture decisions and mentoring team members on AI/ML and clean architecture best practices

Senior Software Developer

Derivco | Nov 2018 - Jun 2025

  • Led technical development of internal AI initiative, creating a dynamic prompt generation system
    • JSON-based prompt definitions interpreted at runtime for customizable, reusable prompts
    • Integrated with OpenAI API for automated content generation workflows
    for OpenAI API integration.
  • Architected LLM-powered automation tools
    • Automated test generation reducing SDET manual effort
    • AI-assisted code review workflows for quality and consistency
    for SDET and code review workflows.
  • Applied iDesign architectural methodology
    • Formal training under Juval Löwy, creator of iDesign
    • Decomposition into Managers, Engines, Resource Accessors, and Utilities
    • Strict layered dependency direction with protocol-based contracts
    to create maintainable, scalable backend systems.
  • Led the technical development, release, and maintenance of end-to-end software systems
    • Full ownership from requirements through design, implementation, testing, and production deployment
    • Coordinated cross-team releases across multiple dependent services
    • Maintained high-availability systems in the online gaming industry
    C# .NET SQL Server REST APIs
    in the online gaming industry.
  • Drove CI/CD pipelines
    • Automated build, test, and deployment workflows
    • Standardised branching strategies and merge policies
    • Integrated automated testing gates to enforce quality before release
    TeamCity Octopus Deploy Git
    and technical roadmap for multiple software products.
  • Assessed complexity, scope, risk, and engineering effort for significant software development projects.
  • Contributed to architecture roadmaps
    • Ensured adherence to iDesign decomposition and layering principles
    • Reviewed and guided system design for new features and services
    • Drove alignment between technical strategy and business objectives
    , ensuring adherence to established standards and principles.
  • Mentored junior developers, providing guidance and support.

Key References:

Senior Software Developer GMIC.Net (Consultant)

Dynamic Visual Technologies (Pty) Ltd | May 2016 - Nov 2018

  • Developed new features and critical bug fixes on the GMIC.Net platform
    • Enterprise .NET application serving the mining and minerals industry
    • Multi-layered architecture with WCF services, business logic, and data access tiers
    • Complex domain logic for commodity tracking, compliance, and reporting
    C# .NET WCF SQL Server
    as per client requirements.
  • Handled continuous integration
    • Managed automated build pipelines on client-hosted infrastructure
    • Configured build triggers, artefact publishing, and deployment steps
    • Maintained build health and resolved pipeline failures
    (build) processes on clients' preferred platform.
  • Managed branches and releases
    • Git-based branching strategy with feature, release, and hotfix branches
    • Coordinated merge workflows across multiple development streams
    • Orchestrated releases to production after QA sign-off
    Git Release Management
    using Git.
  • Collaborated closely with Quality Assurance on quality-related problems and bugs.
  • Managed sprint goals and workload in coordination with project manager, product owner, and team capacity.
  • Executed system integrations
    • Integrated GMIC.Net with third-party systems at client sites
    • Handled data synchronisation, API integration, and cross-system workflows
    between various systems at client sites.
  • Participated in technical discussions and design sessions relating to product development.

Key Reference:

Software Developer/Windows Team Lead

Seecrypt (Pty) Ltd | Apr 2015 - May 2016

  • Developed new features and bug fixes on the Seecrypt SC3
    • Secure encrypted communications application for voice, messaging, and file transfer
    • End-to-end encryption with military-grade security protocols
    • Cross-platform support targeting Windows desktop, phone, and tablet
    Encryption Secure Comms VoIP
    encrypted communications application.
  • Developed with a multi-language tech stack
    • C# for application logic and UI layer
    • C++ for performance-critical cryptographic operations
    • WPF for desktop UI with rich XAML-based interfaces
    • Silverlight for cross-platform web-based client
    C# C++ WPF Silverlight XAML
    including C#, C++, WPF, and Silverlight.
  • Implemented encrypted local storage
    • SQLite with SQLCipher extension for transparent AES-256 database encryption
    • Secure storage of message history, contacts, and session data
    • Key management for per-user encrypted database instances
    SQLite SQLCipher AES-256
    using SQLite-SQLCipher.
  • Managed CI/CD with Bamboo and Stash
    • Bamboo build server for automated build and test pipelines
    • Stash (Bitbucket Server) for Git repository hosting and code review
    • Automated deployment to Windows Store after QA sign-off
    Bamboo Bitbucket Git Windows Store
    , releasing to Windows Store after QA sign-off.
  • Led the Windows team as team lead
    • Managed sprint goals and workload in coordination with scrum master and product owner
    • Drove regular bug-bashes with QA to maintain quality standards
    • Contributed to cross-team design sessions with other team leaders
    • Assisted team members with debugging and new feature development
    Scrum Team Leadership
    , coordinating with scrum master, product owner, and QA.

Key References:

Director

Covariant Consulting (Pty) Ltd | Jun 2014 - Mar 2015

  • Attended and assisted in design meetings regarding the implementation of the UDCS system
    • UD Truck Customer Service system for dealer network management
    • End-to-end solution covering service scheduling, parts ordering, warranty claims, and dealer communication
    C# .NET SQL Server
    .
  • Integrated the UDCS front-ends and back-ends into Syspro ERP
    • Syspro integration for finance, inventory management, and business reporting
    • Data synchronisation between UDCS and Syspro modules
    • Custom reporting solutions bridging both systems
    Syspro ERP Integration SQL Server
    for finance, inventory, and reporting purposes.
  • Implemented database, data-access, business logic, and rules for the full UDCS system.
  • Performed testing and verification of all implemented code on a standalone QA system.
  • Managed deployment and verification of new functionality and bug fixes after QA sign-off.
  • Maintained software and data fixes as needed and provided dealer support.

Key Reference:

Software Developer

UD Trucks Southern Africa (Pty) Ltd | Mar 2012 - May 2014

  • Designed and implemented the UDCS system
    • Core system architecture for dealer customer service operations
    • Full-stack implementation: database schema, stored procedures, data access layer, business rules, and front-end
    • Standalone QA environment for isolated testing and verification
    C# .NET SQL Server
    architecture and integration.
  • Integrated UDCS with Syspro ERP
    • Bi-directional data flow between UDCS and Syspro finance and inventory modules
    • Automated reporting pipelines for business intelligence
    Syspro ERP Integration
    for finance, inventory, and reporting.
  • Managed deployment and verification of new functionality and bug fixes after QA sign-off.
  • Maintained software and data fixes as needed and provided dealer support.

Key Reference:

Software Developer

Enermatics (Pty) Ltd | Jun 2011 - Mar 2012

  • Designed and implemented the Wattkeeper system
    • Energy management platform for monitoring and controlling electrical consumption
    • Real-time data collection from smart meters and energy monitoring devices
    • Web-based dashboard for usage analytics, billing, and reporting
    C# .NET Energy Management IoT
    front-end and back-end, including database, data-access, and business logic.
  • Implemented device communication protocols
    • Serial and TCP/IP communication with energy monitoring hardware
    • Protocol adherence as specified by device engineers
    • Data parsing, validation, and storage of meter readings
    Device Protocols Serial Comms TCP/IP
    for communicating with energy monitoring devices.
  • Performed testing and verification of all implemented code on a standalone QA system.
  • Managed deployment and verification using Hudson build server
    • Automated build and deployment pipelines (Hudson, precursor to Jenkins)
    • Continuous integration for build verification and automated testing
    Hudson CI/CD
    .
  • Assisted with client queries and billing cycles.

Key Reference:

Software Developer

Infotech (Pty) Ltd | Jun 2009 - May 2011

  • Designed and implemented the Wattkeeper system
    • Initial development of the energy management platform later continued at Enermatics
    • Full-stack implementation: database schema, data access layer, business rules, and web front-end
    • Foundation architecture for device data collection and energy analytics
    C# .NET Energy Management
    front-end and back-end, including database, data-access, and business logic.
  • Implemented device communication protocols
    • Communication with energy monitoring hardware per device engineer specifications
    • Data ingestion pipelines for real-time meter readings
    Device Protocols IoT
    as specified by device engineers.
  • Performed testing and verification of all implemented code on a standalone QA system.
  • Managed deployment and verification using Hudson build server
    • Automated build pipelines with Hudson (precursor to Jenkins)
    • Build verification and deployment automation
    Hudson CI/CD
    .

Key Reference:

Programmer

Information Trade World | Feb 2007 - May 2009

  • Managed projects and client relationships.
  • Developed applications in Visual Basic and Delphi
    • Desktop application development with Visual Basic 6 and Delphi
    • Database-driven business applications with MS Access and SQL backends
    • Website development for client-facing web presence
    Visual Basic Delphi Web Development
    , along with website development.
  • Designed, developed, and maintained COBOL programs
    • Mainframe COBOL development for client business systems
    • Batch processing, report generation, and data transformation
    • Full lifecycle: design, development, implementation, and maintenance
    COBOL Mainframe Batch Processing
    as per client requirements.
  • Handled client system management, including scheduling, planning of client jobs, and writing progress reports.

Key Reference:

Administrative Officer

City of Tshwane Metropolitan Municipality | Mar 2004 - Feb 2007

  • Checked and controlled payment vouchers for goods and services, ensuring all legal documents were attached and correct.
  • Ensured contracts with retention monies complied with General Conditions of Contract.
  • Verified correct foreign exchange covers were obtained.
  • Ensured all payments including retention monies were correctly paid in accordance with contract conditions.
  • Provided a counter service to internal departments for cheque collection.
  • Printed cheques as required.
  • Oversaw the overall status of filed documents and managed contents, archiving when necessary.
  • Assisted departments with queries regarding payment status.
  • Assisted with all IT related problems, sometimes resolving them immediately.
  • Performed general administrative services.

Key Reference:

Graphic Designer / Sign Writer

Marnet Signs | Nov 2003 - Feb 2004

  • Created signboards and vehicle signage with graphical processing software.
  • Incised vinyl for signs using a graphical plotter.
  • Weeded superfluous vinyl from backing sheets.
  • Applied vinyl onto various boards and vehicles to create the actual sign.
  • Ordered materials required for projects as necessary.
  • Obtained materials from various suppliers with personal vehicle.
  • Oversaw and helped with the actual erection of signage.

Key Reference:

Data Capturing Clerk

Books-u-profit | Apr 2003 - Jul 2003

  • Performed general administration services (filing, arranging appointments, etc.).
  • Captured all invoices, receipts, and financial documents of clients onto a computer, using Pastel 6 accounting software.

Key Reference:

Graphic Designer / Sign Writer

Nexor Signs | Dec 2002 - Mar 2003

  • Created signboards and vehicle signage with graphical processing software.
  • Incised vinyl for signs using a graphical plotter.
  • Weeded superfluous vinyl from backing sheets.
  • Applied vinyl onto various boards and vehicles to create the actual sign.
  • Assisted with actual erection of signage.

Key Reference:

Diploma in C++ Programming

Intec College

Maintained an average of 73.1%, the highest score obtained in the history of the institution as of 2006.

Subjects:

  • Introduction to programming
  • Introduction to C++
  • C++ Programming
  • Advanced C++ Programming
  • Systems Analysis & Design
  • Management Information Systems.

Grade 12 (Matric with Merit)

Goudrif High and Elsburg Primary

Subjects:

  • Afrikaans: First Language HG - C
  • English: Second Language HG - A
  • Physical Science - D
  • Accounting - C
  • Computer Studies - B
  • Mathematics - E

Recipient of the Computer Training Institute floating trophy for the best student in computer studies (2002).

Completed entire Matric Computer Studies syllabus in Grade 10 and coded first primitive graphics program in the same year.

Programming Languages

C# & .NET 10 Ecosystem

Primary specialty. Architecting and developing high-performance, scalable back-end services, APIs, and AI platform frameworks. Deep expert-level proficiency from legacy .NET Framework through to .NET 10.

Python (LangChain, LangGraph, FastAPI)

Building agentic chatbot platforms and RAG orchestration services with LangChain/LangGraph for LLM orchestration, FastAPI for backend APIs, and Pydantic for data validation.

TypeScript / React

Building modern frontend interfaces for AI platforms, admin dashboards, and real-time chat applications with streaming support.

Dart / Flutter

Cross-platform mobile and desktop chat interfaces with real-time SSE/WebSocket streaming integration.

SQL (T-SQL/General)

Designing and implementing complex database schemas, writing optimized queries, and ensuring data integrity for mission-critical systems.

C

Native terminal applications with ncurses and SQLite, including tawk, an open source WhatsApp client for the terminal.

C++

Performance-critical components in secure communications applications, interfacing directly with low-level libraries and protocols.

Legacy & Other (COBOL, Delphi, VB)

Foundational experience in maintaining and developing for legacy systems, demonstrating broad adaptability across programming paradigms.

Frameworks & Methodologies

AI/ML Engineering

LangChain, LangGraph, Model Context Protocol (MCP), Agent-to-Agent (A2A) protocol, RAG pipelines
  • 14 chunking strategies: recursive, semantic, sentence, fixed, markdown, HTML, code, table-aware, hierarchical, contextual, proposition, late, agentic, compression
  • Hybrid retrieval combining BM25 keyword matching with dense vector search
  • Score fusion via Reciprocal Rank Fusion (RRF) and weighted strategies
  • RAPTOR hierarchical summarization for multi-level context understanding
  • HyDE (Hypothetical Document Embeddings) for semantic intent matching
  • ColBERT late interaction for fine-grained token-level relevance
  • Graph RAG for relationship-aware retrieval
  • Deep search patterns for multi-step retrieval
  • Trustworthy generation with confidence scoring and hallucination detection
  • Fluent builder API with presets: Conversational, Production, High-Accuracy, Code Search, Research
  • Query transformation and expansion for improved recall
  • Document ingestion pipeline with metadata extraction and versioning
, agentic orchestration
  • Planning coordinator with 3 strategies: ReAct (iterative reasoning), Plan-Execute (upfront planning), Adaptive (LLM auto-selects based on task complexity)
  • Tool registry with automatic discovery, execution with retry logic, and canary token support
  • Multi-model routing by engine type: agentic, planning, reflection, classification
  • LLM provider registry for runtime model selection and fallback
  • MCP tool loader for external tool server composition via Model Context Protocol
  • Vector-backed conversation memory with semantic similarity search
  • Structured output generation with retry logic and validation steps
, AI security
  • Agent security monitor with risk accumulation and decay over conversation turns
  • Multi-turn attack detection tracking escalation patterns across conversation history
  • Risk scoring with configurable thresholds triggering capability restriction or human escalation
  • Prompt injection scanning with pluggable scanner adapter registry
  • Canary token management for data exfiltration detection
  • Tool chain validation preventing unauthorized operation sequences
  • Input sanitization and spotlighting (user input wrapped in security delimiters)
  • Output filtering detecting system prompt leakage patterns
, context management
  • Rolling window manager with sliding window over conversation history
  • Fallback summarization when context exceeds window limits
  • Context assembly for RAG results with relevance-based selection
  • Conversation reranking and compression for long conversations
, Vector Databases (Qdrant, FAISS, Pinecone, ChromaDB, Weaviate), Prompt Engineering, Semantic Search, Embeddings, multi-agent systems
  • Hierarchical agent orchestration with recursive depth control
  • Up to 37 specialized reviewers covering backend (8 languages) and frontend (4 frameworks)
  • Proof-gated verification: an independent service re-runs acceptance checks before work is accepted
  • Human-in-the-loop design: approval gates, two-step confirmation, untrusted text fencing
  • Master orchestrators coordinating review and fix agents
  • Bounded scope per agent preventing context fragmentation
, A/B Testing.

Multi-Protocol Backend Architecture

Designing and implementing APIs across gRPC, SignalR, Server-Sent Events (SSE), Streamable HTTP, WebSocket, and REST. Building multi-transport server frameworks that support session management, authentication, and automatic tool discovery.

Frontend & Mobile

React, TypeScript, Flutter, Dart, Tailwind CSS. Building real-time streaming interfaces for AI chat platforms across web and mobile.

DevOps & Infrastructure

Docker, Docker Compose, Rancher, Kubernetes, Azure, CI/CD Pipelines, cross-platform containerized deployments (Linux, Mac, Windows). Driving continuous integration and delivery pipelines using Azure DevOps.

Observability & Metrics

OpenTelemetry, Elasticsearch, Kibana, structured logging with correlation IDs, A/B testing frameworks
  • Experiment and variant management with assignment tracking
  • Metric collection and analysis for model evaluation
  • Side-by-side response comparison for provider benchmarking
.

Testing

NUnit, xUnit, pytest, Vitest, integration testing, architecture and convention testing, coverage gates in CI. Building test generation tools powered by LLMs to reduce QA burden.

Architectural Concepts

iDesign Methodology & Domain-Driven Design (DDD)

Formally trained in advanced system architecture through Juval Löwy's prestigious Architect's Master Class. Proficient in applying the iDesign methodology and Domain-Driven Design principles to decompose complex systems into robust, maintainable, and scalable software.

Clean Architecture

Structuring AI platforms and backend services with clear separation of concerns, dependency inversion, and testable domain logic independent of frameworks and infrastructure.

Microservices & Event-Driven Architecture

Architecting and decomposing systems into scalable, independently deployable microservices with event-driven communication patterns, leveraging Azure for orchestration and deployment.

Authentication & Security

JWT and API key authentication, multi-provider auth
  • Azure AD, JWT, and API key authentication providers
  • Microsoft Entra ID credential strategies with a coalescing token refresh service
  • RBAC engine for role-based access control
  • Token validation and middleware integration
, and security-first design for AI platform APIs and MCP server frameworks.

Agentic AI Design Patterns

Implementing production-grade agentic security patterns
  • Spotlighting: user input wrapped in security delimiters to prevent prompt injection
  • Input/output guardrails with sanitization and leakage detection
  • Multi-turn attack scoring with risk accumulation and decay across conversation turns
  • Canary token injection for data exfiltration detection
  • Tool chain validation preventing unauthorized operation sequences
  • Human-in-the-loop escalation when risk thresholds are exceeded
  • Capability restriction dynamically reducing agent permissions under threat
and orchestration patterns
  • ReAct: iterative Thought-Action-Observation reasoning loops
  • Plan-Execute: upfront planning followed by step-by-step execution
  • Adaptive: LLM auto-selects strategy based on task complexity
  • Tool use governance with validation boundaries and retry logic
  • Structured output generation with schema validation and retry
  • Content style enforcement for consistent LLM output formatting
for enterprise AI platforms.

Database & Vector Storage

SQL databases (SQL Server, PostgreSQL, SQLite), vector databases (Qdrant, FAISS, Pinecone, ChromaDB, Weaviate), caching (Redis, in-memory), and optimized data access patterns for hybrid search systems.

Open Source

Software I build in the open and am proud of. The code is on GitHub.

Open Source

tawk: WhatsApp in the Terminal

An open source WhatsApp client for the terminal, written in C with ncurses. It links to your phone the same way WhatsApp Web does, keeps chats in a local SQLite database, and brings notifications, voice notes with transcripts, media, statuses and themes to the terminal. AI agents can work in it under rules you set chat by chat. Independent project, not affiliated with WhatsApp or Meta.
  • Two interchangeable WhatsApp backends behind one contract, switched with one config key
  • Up to eight accounts in one window, with photos and PDFs drawn as real pixels on Sixel terminals
  • Voice notes transcribed in the language spoken, and a TL;DR mode that folds long messages into summaries written by a connected agent
  • An owner’s chat: talk to your agent and approve its requests from WhatsApp on your phone
  • Agent rules per chat (always ask, read only, hidden), with one-time codes and card numbers masked before a model sees them
  • Encrypted chats, tamper-evident backups and a lock that asks for your passphrase
  • Labels, reminders and filters for a long chat list, and system notifications with quiet hours
tawk showing a group chat with a photo, reactions and an edited message
C ncurses SQLite SQLCipher Go Sixel MIT
github.com/loganventer/tawk
Open Source

tawk-mcp: MCP Server for tawk

An open source Model Context Protocol server that lets AI clients such as Claude Code, Claude Desktop and VS Code read WhatsApp chats through a running tawk and act on them with your approval. Every send waits for approval in tawk, and anything destructive needs two separate confirmations, one of which the model can never give.
  • More than 90 tools across chats, messages, scheduling, statuses, labels, reminders, voice notes and memory
  • Six iDesign layers with every interface bound in one composition root
  • Message text reaches the model inside clearly marked blocks it is told not to obey
  • Writes voice note transcripts and TL;DR summaries when tawk asks for them
  • Hears you in your own chat as instructions, and treats everything else as data
  • Memory for writing like you: voices tuned per audience, contact profiles and reply templates
  • Memory syncs between your machines through a private git repository of your own, on a schedule or when an agent asks
graph LR
  Client[AI Client] -->|MCP| Server[tawk-mcp]
  Server -->|Control socket| Tawk[tawk]
  Tawk --> You[Your Approval]
  Tawk --> WA[WhatsApp]
  style Client fill:#1e3a5f,stroke:#3b82f6,color:#93c5fd
  style Server fill:#1e293b,stroke:#475569,color:#94a3b8
  style Tawk fill:#1e3a5f,stroke:#3b82f6,color:#93c5fd
  style You fill:#1e3a5f,stroke:#3b82f6,color:#93c5fd
  style WA fill:#1e3a5f,stroke:#3b82f6,color:#93c5fd
.NET 10 C# MCP C# SDK ASP.NET Core SQLite Docker MIT
github.com/loganventer/tawk-mcp
Open Source

matrix-clock: Terminal Clock in Digital Rain

A terminal mashup of cmatrix and tty-clock: the Matrix digital rain falls down the screen and renders a large digital clock, either punched out of the rain as a silhouette or filled with falling code. A task panel above the clock shows the running task from Toggl Track and Azure DevOps.
  • Four clock styles, seven rain colours, adjustable size and speed
  • Live task lines with a full-screen browser of the day’s entries
  • A stats overlay with frame time, frames per second and CPU usage
matrix-clock showing the time as a silhouette in green digital rain
C ncurses libcurl Assembly

Personal Interests & Projects

Software Development

Passionate about programming, finding it exhilarating and rewarding to solve particularly hard problems.

Tech Innovations

  • Builds open source terminal software: tawk, a WhatsApp client for the terminal, tawk-mcp, its MCP server, and matrix-clock, a digital rain clock. View projects →
  • Developed a mobile game for the Windows phone store which received 8000 downloads.
  • Developed another Windows phone app for the Eskom load shedding situation, monitoring grid status and notifying users of changes, which received 1200 downloads in two weeks.
  • Built a backlight for his television which changes color according to the average perceived pixel color, greatly improving the viewing ambience.

Music

Highly interested in music, plays the guitar and sings at home and in public. Has written many of his own songs and performed at various venues after one of his songs was played on Jacaranda, and recorded his own album.

AI Agent Demo

An AI agent you can chat with about my experience, skills, and projects. Powered by Claude.

Powered by Claude

All References

Derivco

Chris Coetzee: +27 82 548 8817

Ockert Du Preez: +27 72 829 0170

Paul Kotze: +27 82 901 7953

Albie van Zyl: +27 79 492 5393

Gopi Kistasami: +27 83 616 1371

Dynamic Visual Technologies

Khumo Jele: +27 81 447 9732

Seecrypt (Pty) Ltd

Annalinde Singh: +27 76 640 4766

Dino Fizzotti: +44 734 041 5199

UD Trucks / Covariant Consulting

Ronete McKenzie: +27 61 682 2350

Infotech/Enermatics

Melissa Wittstock: +27 82 750 0065

Information Trade World

Maria Jacobs: +27 74 254 6027

City of Tshwane

Martie J. Matthee: +27 12 358 8269

Marnet Signs

Martie Roberts or Eddie Bekker: +27 12 734 4990

Books-u-profit

Dries Venter: Tel: +27 12 800 3327, Cell: +27 83 449 7047

Nexor Signs

Dries du Plessis: +27 12 333 3384