I build AI systems that are auditable, not just capable, multi-agent knowledge graphs, VLM pipelines, and software for social impact.
Currently: SF 🌁 · MIND Lab · Hack4Impact engineering
| Project | What it does | Stack |
|---|---|---|
| 🦺 IronSite | VLM construction safety pipeline — adversarial CoT, OSHA mapping, re-ID across shifts | YOLO11, SAM3, Qwen3-VL, LoRA, FastAPI |
| 🧠 MAGG | Autonomous multi-agent knowledge base with adversarial cross-validation | LangChain, LangGraph, Ollama, HuggingFace |
| 🤝 For All Ages | AI matching: Pinecone embeddings + Hungarian algorithm for 200+ users | React, Pinecone, PostgreSQL, Firebase |
| 🐝 Beta-Tester-Swarm | 12 persona-diverse AI agents QA-test mobile apps on parallel iOS simulators — outputs severity-ranked defect report + Swarm Rating | Next.js, TypeScript, LLM Vision, lim.run, Replicas |
| 🚀 ATOMZ | Multi-agent Mars construction RL env — LangGraph orchestration Brain dispatches a robot fleet through a MuJoCo physics sim; pi0.5 VLA finetuned with RL | Python, MuJoCo, LangGraph, Modal, HUD SDK |
AI/ML: LangChain LangGraph HuggingFace Ollama YOLO SAM LoRA Pinecone RAG spaCy scikit-learn
Backend: FastAPI Express GraphQL Prisma REST Flask
Cloud: AWS Lambda API Gateway S3 SNS SES DynamoDB EventBridge Amplify
Data: PostgreSQL Firebase MongoDB SQLite DynamoDB Vector DBs Knowledge Graphs
| Paper | Venue | Year |
|---|---|---|
| Foundations and Frontiers of Multimodal Agentic Frameworks | TMLR | 2026 |
| IronSite: Construction Safety VLM System | arXiv | 2026 |
| MAGG: Multi-Agent Governed Knowledge Graph | EMLNP (submitted) | 2026 |
Thanks for scrolling 🌙 · always building, always shipping