Engineering Complexity.
Shipping Systems That Work.
I build production AI agents, enterprise architecture platforms, and quant finance systems โ spanning healthcare graph-RAG, blockchain payments, autonomous trading, and LLM-native ERP. Every project below runs in production or active development.
ClinicalMatch AI
Precision clinical trial matching powered by FHIR R4 and GraphRAG
A full-stack AI agent that matches patients to recruiting clinical trials using a Neo4j knowledge graph, real-time ClinicalTrials.gov data, and FHIR R4 patient profiles. Built for the "Agents Assemble: Healthcare AI Endgame Challenge" and deployed live on HuggingFace Spaces.
- โFHIR R4 patient ingestion โ SMART on FHIR & synthetic profiles
- โNeo4j knowledge graph: 500 patients ยท 250+ trials ยท 9,100+ ELIGIBLE_FOR edges
- โGraphRAG natural language queries over the clinical graph
- +6 more capabilities
500
Synthetic patient profiles in graph
250+
NCT trials indexed
9,100+
ELIGIBLE_FOR graph edges
6
MCP tools for AI agent integration
What's included
- โFHIR R4 patient ingestion โ SMART on FHIR & synthetic profiles
- โNeo4j knowledge graph: 500 patients ยท 250+ trials ยท 9,100+ ELIGIBLE_FOR edges
- โGraphRAG natural language queries over the clinical graph
- โA2A 5-state orchestration pipeline (INGEST โ PARSE โ MATCH โ SCORE โ RECRUIT)
- โMCP server with 6 callable tools for AI agent integration
- โRecruitment Hub: Kanban board tracking patients from IDENTIFIED โ ENROLLED
- โPersonalized AI outreach generation (PCP letter, patient email, social post)
- โReal-time ClinicalTrials.gov v2 integration โ NCT data auto-ingested into graph
- โDeployed: Docker multi-stage build, Nginx reverse proxy, Supervisord, HuggingFace Spaces
Job Hunter KE
AI-powered job application automation for the Kenyan market
A full-stack SaaS that scrapes Kenyan and global remote job boards, generates tailored CVs and cover letters with AI, routes them through a React dashboard for review, and dispatches approved applications. Runs on Flask + n8n + SQLite with Supabase auth.
- โ14 job board scrapers โ LinkedIn, BrighterMonday, Fuzu, RemoteOK, Himalayas, WeWorkRemotely + specialist quant boards
- โAI document generation โ tailored CV + cover letter per job via Groq / OpenRouter
- โn8n workflow orchestration โ scrape โ generate โ email โ dispatch pipeline
- +5 more capabilities
14
Job board scrapers integrated
3
AI providers supported
5
n8n workflow automations
Live
Running in production daily
What's included
- โ14 job board scrapers โ LinkedIn, BrighterMonday, Fuzu, RemoteOK, Himalayas, WeWorkRemotely + specialist quant boards
- โAI document generation โ tailored CV + cover letter per job via Groq / OpenRouter
- โn8n workflow orchestration โ scrape โ generate โ email โ dispatch pipeline
- โReact dashboard with sortable job table, status tracking, and document preview
- โEmployer reply inbox with AI-powered response drafting
- โSupabase auth with Google OAuth and magic link
- โBackground task runner with ThreadPoolExecutor and SQLite task log
- โScholarship scraper module running in parallel with job pipeline
GraphAlpha
Autonomous multi-agent trading system with knowledge graph signal generation
A multi-agent autonomous trading system combining a Neo4j/Memgraph knowledge graph for signal generation with 8 specialised sub-agents โ regime classification, news sentiment, macro calendar, derivatives pricing, risk management, and execution routing to IBKR and Kraken.
- โ8 specialised sub-agents: Regime ยท Signal ยท News ยท MacroCalendar ยท KGSignal ยท KGDerivatives ยท Risk ยท Execution
- โKnowledge graph formula evaluation โ KG nodes encode trading signals evaluated per tick
- โGARCH volatility modelling + VARLiNGAM causal discovery (pgmpy, lingam)
- +6 more capabilities
8
Autonomous sub-agents
2
Execution venues (IBKR + Kraken)
GARCH
Volatility model
Paper
Current trading mode
What's included
- โ8 specialised sub-agents: Regime ยท Signal ยท News ยท MacroCalendar ยท KGSignal ยท KGDerivatives ยท Risk ยท Execution
- โKnowledge graph formula evaluation โ KG nodes encode trading signals evaluated per tick
- โGARCH volatility modelling + VARLiNGAM causal discovery (pgmpy, lingam)
- โOptions pricing engine: Black-Scholes, Heston SV, FFT, live Greeks
- โDual execution: IBKR TWS (equity + options) and Kraken (crypto perps)
- โRedis inter-agent messaging bus with async orchestration loop (5-min ticks)
- โPrometheus metrics: PnL gauge, drawdown gauge, signal count, loop counter
- โHard risk limits: max drawdown halt, position sizing via RiskAgent
- โPaper trading mode with full audit trail before live deployment
Agentic ERP
AI-native enterprise resource planning in Go with 27 MCP tools
A full-stack, AI-native ERP platform where every module is driven by a provider-agnostic LLM layer. Users describe intent in natural language; the AI selects the right ERP tool and drafts actions for human approval before any write occurs. Ships as both a web app and a native desktop app (Wails v3).
- โ6 ERP modules: Financial (GL/AP/AR) ยท HR ยท CRM ยท Supply Chain ยท Project Mgmt ยท AI Agent
- โ27 MCP tools in the registry โ full ERP function-calling surface for LLMs
- โProvider-agnostic LLM client โ Groq default, OpenAI-compat, Anthropic, Ollama
- +7 more capabilities
27
MCP tools in registry
6
ERP service modules
2
Deployment targets (web + desktop)
Go
Core language
What's included
- โ6 ERP modules: Financial (GL/AP/AR) ยท HR ยท CRM ยท Supply Chain ยท Project Mgmt ยท AI Agent
- โ27 MCP tools in the registry โ full ERP function-calling surface for LLMs
- โProvider-agnostic LLM client โ Groq default, OpenAI-compat, Anthropic, Ollama
- โConnectRPC inter-service communication with Protobuf definitions
- โNATS JetStream event bus for domain events + IoT leaf nodes
- โTemporal for crash-safe long-running workflow orchestration
- โKeycloak OIDC + Casbin RBAC/ABAC for enterprise auth
- โClean Architecture across all services (domain โ application โ infrastructure โ interfaces)
- โDual deployment: Next.js 15 web app + Wails v3 desktop app (same frontend)
- โGCP Terraform infra + Kubernetes Helm charts per service
AMD EA Strategy Optimizer
Enterprise Architecture intelligence powered by AMD MI300X, GraphRAG, and DRL
An AI-native Enterprise Architecture platform built for the AMD Developer Hackathon 2026. Transforms business goals into governance-grounded, Jira-ready strategic roadmaps using a 1,416-capability Neo4j knowledge graph, Deep Reinforcement Learning prioritisation, and a self-correcting LangGraph agentic pipeline โ all served from AMD Instinct MI300X via vLLM.
- โNeo4j knowledge graph: 44 domains ยท 248 subdomains ยท 1,416 capabilities ยท 200+ trends
- โLangGraph 4-node agentic pipeline: Retrieve โ Optimize โ Generate โ Verify (self-correcting)
- โDeep Reinforcement Learning prioritisation โ MLP trained on governance reward signals
- +6 more capabilities
1,416
EA capabilities in knowledge graph
44
Enterprise domains modelled
MI300X
AMD GPU (192 GB HBM3)
Jira
Live export integration
What's included
- โNeo4j knowledge graph: 44 domains ยท 248 subdomains ยท 1,416 capabilities ยท 200+ trends
- โLangGraph 4-node agentic pipeline: Retrieve โ Optimize โ Generate โ Verify (self-correcting)
- โDeep Reinforcement Learning prioritisation โ MLP trained on governance reward signals
- โAMD Instinct MI300X inference: Qwen2.5-72B at fp16, 192 GB HBM3, SSE streaming
- โStrategic Roadmap generator: questionnaire โ Epics โ Features โ User Stories โ Tasks
- โLive Jira REST API v3 export + ServiceNow / Azure DevOps integration
- โGraph Explorer: interactive force-directed network of 44 EA domains
- โChat session persistence backed by Neo4j โ full conversation memory per project
- โExport handover: JSON / Markdown / CSV of complete roadmaps
Lex Kenya
Constitutionally-anchored GraphRAG intelligence over Kenyan corporate and statutory law
A proprietary GraphRAG legal intelligence platform built over a Neo4j knowledge graph of Kenyan law โ the Constitution of Kenya 2010, 19 Acts, 997 court judgments, and 28 legal concepts. Every answer is traceable to its constitutional authority chain via graph traversal. Targets corporate lawyers, in-house counsel, compliance officers, and legal researchers operating in the Kenyan jurisdiction.
- โNeo4j knowledge graph: 4,862 nodes ยท 8,417 edges โ Constitution, 19 Acts, 997 judgments
- โConstitutional authority chain: every section traceable to its Article via DERIVES_AUTHORITY_FROM
- โGraphRAG Q&A โ Cypher traversal + BGE-M3 semantic retrieval, answers cite specific legal sources
- +6 more capabilities
4,862
Knowledge graph nodes
8,417
Graph relationships
997
Court judgments indexed
19
Acts + Constitution ingested
What's included
- โNeo4j knowledge graph: 4,862 nodes ยท 8,417 edges โ Constitution, 19 Acts, 997 judgments
- โConstitutional authority chain: every section traceable to its Article via DERIVES_AUTHORITY_FROM
- โGraphRAG Q&A โ Cypher traversal + BGE-M3 semantic retrieval, answers cite specific legal sources
- โCompliance Checker โ input a business action, get relevant statutory constraints with constitutional basis
- โRuling Predictor โ precedent analysis across 997 judgments from 5 superior courts
- โQLoRA fine-tuned Mistral-7B on 700 Kenyan law QA pairs (Unsloth + TRL)
- โCase law scraper โ kenyalaw.org with court-specific pagination logic
- โGraph Admin panel โ live stats, scraper controls, node/relationship counts
- โ28 legal concepts modelled (Rule of Law, Transfer Pricing, etc.) with constitutional grounding
Quantifaya Content
Interactive quant finance research โ live charts, LaTeX, and Jupyter notebooks
A free research hub publishing quantitative finance deep-dives with interactive Plotly charts, LaTeX-rendered mathematics, and executable Jupyter notebooks. Two series: Classical Quantitative Finance and DeFi Mechanics.
- โ9 interactive chart components: Efficient Frontier ยท Correlation Heatmap ยท VaR ยท Factor Model ยท AMM Curve + more
- โLaTeX-rendered mathematics via KaTeX โ full equation support inline and block
- โJupyter notebook access gate โ Colab + download for every episode
- +4 more capabilities
9
Interactive chart components
2
Structured series
Free
All posts and charts
Live
New episodes published regularly
What's included
- โ9 interactive chart components: Efficient Frontier ยท Correlation Heatmap ยท VaR ยท Factor Model ยท AMM Curve + more
- โLaTeX-rendered mathematics via KaTeX โ full equation support inline and block
- โJupyter notebook access gate โ Colab + download for every episode
- โ2 structured series: Classical Quantitative Finance + DeFi Mechanics
- โMDX-powered posts with syntax highlighting, copy buttons, and custom components
- โSupabase-backed subscriber and lead management with Resend email delivery
- โTopics: portfolio variance, efficient frontier, Sharpe/Sortino, VaR, factor models, AMMs
Latest from the Blog
Animated videos, deep-dive articles, and executable code โ learn quant finance by doing
How Uniswap Actually Works: The xยทy=k Formula
Every stock exchange uses order books and human market makers. Uniswap replaced all of that with one equation: xยทy=k. Derives the constant product formula and calculates price impact from first principles.
Why Correlation Matters More Than Returns
Adding a losing asset to your portfolio can make you more money. Derives the portfolio variance formula, builds a correlation matrix from scratch in Python, and shows why diversification fails exactly when you need it most.
Open to Opportunities
Let's build something together
I'm available for full-time, contract, and remote roles in Full-Stack Engineering, Solutions Architecture, AI/ML, and Financial Engineering. Based in Nairobi โ open globally.