Leeon John
I build production-oriented AI systems focused on reliability, evaluation, agent orchestration, and developer tooling.
OpenSourcePilot
AI-powered open-source contribution assistant that analyzes GitHub repositories, understands issues, performs semantic code search, generates implementation plans, drafts pull requests, and creates test suites.
$ pilot analyze https://github.com/org/repo/issues/42
✓ Repository cloned and indexed
✓ 847 files vectorized in ChromaDB
✓ Issue understood: Add pagination support
→ Searching relevant files…
Found: src/api/routes.py (0.94)
Found: src/models/pagination.py (0.89)
→ Generating implementation plan…
→ Writing test suite…
✓ Pull request drafted (PR #43)
$ ▌
Tech Stack
Agent Workflow
GitHub Issue
Ingests any GitHub issue URL and parses the problem statement
Repository Analysis
Clones and maps the codebase structure, dependencies, and patterns
Semantic Search
ChromaDB vector search finds the most relevant files and functions
Contribution Planning
Planning agent drafts a step-by-step implementation strategy
Test Generation
Auto-generates test suites covering edge cases and happy paths
Pull Request Drafting
Produces a complete, ready-to-submit PR with description and code changes
Systems I've Built
Click any project to explore its architecture, challenges, and engineering decisions.
What I'm Shipping
Active projects in production, staging, and active development.
OpenSourcePilot
AI-powered open-source contribution assistant that analyzes GitHub repos, understands issues, generates implementation plans, and drafts pull requests autonomously.
Current Focus
- ›Semantic search v2
- ›Multi-repo analysis
- ›PR validation loop
ResearchPilot MCP
Agent workflows powered by the Model Context Protocol — giving research agents structured access to tools, memory, and external knowledge sources.
Current Focus
- ›MCP server implementation
- ›Tool registry design
- ›Agent integration
Areas of Expertise
The domains where I design, build, and ship AI systems.
Agentic AI Systems
Designing multi-agent pipelines with Planner, Executor, Critic, and Retriever roles. Autonomous retry loops, structured outputs, and reliable task decomposition.
MCP Infrastructure
Building Model Context Protocol servers and clients that expose structured tools to AI agents — enabling reliable, composable agentic workflows.
LLM Evaluation
Systematic evaluation using LLM-as-judge rubrics, benchmark suites, and iterative self-refinement. Scoring consistency, factuality, and reasoning quality.
AI Reliability
Hardening AI pipelines against hallucinations, prompt injection, and silent failures. Structured outputs, validation gates, and critic-loop feedback.
Retrieval Systems
Building production-grade RAG pipelines with chunking strategies, embedding models, and retrieval ranking. Hybrid search combining dense and sparse methods.
Semantic Search
Vector similarity search across codebases, documents, and knowledge graphs. Fine-tuning retrieval with re-ranking and contextual compression.
Developer Tooling
Building ergonomic CLI tools, APIs, and dashboards that make complex AI systems accessible and observable for engineering teams.
AI Deployment
Containerizing and deploying AI systems with Docker, managing async workloads with FastAPI, and monitoring inference latency and cost.
Under the Hood
How the systems are actually designed. Hover nodes for detail.
OpenSourcePilot
End-to-end agent pipeline from GitHub issue to pull request
GitHub API
Issue & repo ingestion
Repository Analysis
Codebase understanding
Vector Search
ChromaDB semantic retrieval
Planning Agent
Implementation strategy
Test Generation
Automated test suites
PR Drafting
Ready-to-submit pull request
Tools & Technologies
The full spectrum of tools I use to design and deploy AI systems.
LangChain
Agent orchestration & chains
OpenAI API
GPT-4o, embeddings
OpenRouter
Multi-model routing
Anthropic
Claude models
LlamaIndex
Data framework for LLMs
Full Stack
Open Source Work
Building in public at @coderleeon
Let's Build Something
Open to AI engineering roles, interesting projects, and collaborations on agentic systems, MCP infrastructure, and evaluation platforms.