AI Systems Engineer - Agentic AI · Voice · RAG · MLOps · Full-Stack · VoIP
I design and ship production-grade AI systems end-to-end. Six production deployments, real users, built from scratch.
| Area | Technologies |
|---|---|
| AI & Agents | LangChain LangGraph FastMCP CrewAI Anthropic Claude Azure OpenAI OpenAI Hugging Face Prompt Engineering Human-in-the-loop Agent governance A2A flows |
| Voice & Speech | Whisper STT Chatterbox TTS Piper TTS Silero VAD LiveKit FreeSWITCH VoIP/SIP SIP/RTP WebSocket audio streaming Linphone |
| RAG & Search | Qdrant Weaviate ChromaDB Jina Embeddings Crawl4AI SearXNG hybrid search query rewriting multi-source RAG |
| MLOps | Azure ML MLflow continuous retraining CI/CD quality gates model observability automated promotion GPU compute |
| ML & Deep Learning | PyTorch TensorFlow LSTM YOLOv11/12 fine-tuning NLP Behavioural ML |
| Computer Vision | MediaPipe FaceMesh OpenCV YOLO object detection solvePnP head pose estimation |
| Backend & APIs | FastAPI Node.js Express.js REST APIs WebSockets Redis pub/sub Celery BullMQ GraphQL JWT/OAuth |
| Frontend | React Next.js TypeScript TailwindCSS CSS3 Component Architecture |
| Mobile | React Native NativeWind Kotlin |
| Infrastructure | Kubernetes (AKS) Docker NGINX Azure Container Apps GitHub Actions Vercel Linux |
| Cloud | Microsoft Azure IBM Cloud Vercel |
| Databases | PostgreSQL MySQL MongoDB Redis Qdrant Weaviate ChromaDB |
| Languages | Python TypeScript JavaScript Go Kotlin C/C++ Rust Java |
| Network & Telephony | MikroTik RouterOS Yeastar PBX FreeSWITCH SIP trunk administration IVR VLAN segmentation WireGuard VPN |
| Project | Description | Stack |
|---|---|---|
| Elimika AI Tutor | Adaptive agentic learning platform. LangGraph pipeline with real-time multi-source RAG and a custom PyTorch LSTM tracking per-student learning patterns. 1,500+ students, 100% uptime. KCA University pilot incoming. | Python LangGraph PyTorch Qdrant Azure ML MLflow Kubernetes |
| Elimika 365 | Real-time AI exam supervision system. YOLOv11/12 + MediaPipe FaceMesh 478-landmark pose estimation + custom LSTM anomaly model with per-student calibration. 500 concurrent sessions on AKS. | Python FastAPI YOLO PyTorch Redis Azure ML Kubernetes |
| Voicekit | Self-hosted voice agent infrastructure package. Eliminates $0.15–0.25/min platform fees. Whisper STT, Piper TTS, Chatterbox TTS with voice cloning, Silero VAD, OpenAI and Claude LLM routing. True end-to-end token-to-audio streaming. | Python Whisper Chatterbox Silero VAD FastAPI Docker |
| Sieve | VoIP AI call interception on FreeSWITCH. LangGraph agent handles or escalates live calls autonomously. Escalated calls pushed to a Kotlin Android app via FCM over Linphone SIP. Full human-in-the-loop telephony. Zero per-minute costs. | Python FastAPI LangGraph FreeSWITCH Kotlin Redis Docker |
| Prepwise | AI job search operating system. Four LangGraph agents — job search, document generation, interview prep, application tracker — sharing state over WebSockets. Ingests a CV, scores jobs, generates tailored resumes and cover letters. | Python LangGraph Qdrant Jina Claude PostgreSQL Next.js |
| GitHub PR Review MCP Server | FastMCP server exposing GitHub pull request operations as deterministic LLM-callable tools. List PRs, read diffs, post inline comments, approve, request changes. Tested end-to-end with MCP Inspector and Claude Desktop. | Python FastMCP httpx Docker |
| Docker Container Manager MCP Server | FastMCP server wrapping the Docker SDK. Gives any LLM client the ability to list, start, stop, inspect, and monitor Docker containers with accurate CPU and memory stats matching docker stats output. |
Python FastMCP Docker SDK |
| OpenSearch Engine | Zero-cost self-hosted search infrastructure for AI agents. SearXNG metasearch + Spider-rs content extraction, intent classification, parallel engine routing, structured LLM-ready JSON with token counts and relevance scores. | Go Rust Python SearXNG gRPC Docker |
| Events Zawadi | Full-stack social gifting platform with concurrent transaction handling, BullMQ async job queues, and a surprise-preservation mechanic requiring careful state management at the database layer. 100% uptime. | Node.js Express Next.js TypeScript PostgreSQL BullMQ Redis Azure |
Portfolio · LinkedIn · GitHub · koechfestus003@gmail.com · +254 701 183 935