I'm Egor Folley, Forward-Deployed AI Engineer (FDE) at Capco in New York. I help enterprises solve real business problems with AI systems, from C-suite discovery to production deployment.
An open, production-grade reference stack for enterprise agents: 1 shared platform, 3 verticals. Design docs and ADRs land first, code follows. Progress updates weekly (under reconstruction now)
| Project | What it is | Status | Target |
|---|---|---|---|
| AgentGate | Control plane that authenticates, routes, rate-limits and audits every agent and MCP tool call, backed by a permissions knowledge graph | 🟡 Design | Oct 2026 |
| EvalForge | Agents mine an enterprise KG to generate golden sets and red-team cases, then run regression evals over MCP | ⚪ Planned | Nov 2026 |
| OpenDiligence | Multi-agent diligence over SEC EDGAR filings in a company KG. Public rebuild of my PE deal-evaluation pattern, no client IP | ⚪ Planned | Nov 2026 |
| RegGraph | Maps SEC, FINRA and EU AI Act changes onto a firm's policy and controls graph | ⚪ Planned | Dec 2026 |
| WarRoom | Multi-agent incident commander: triage and RCA over a service-dependency graph, human-approved rollbacks | ⚪ Planned | Dec 2026 |
Every repo ships with: LangGraph multi-agent orchestration · Neo4j GraphRAG · MCP servers · OTel/LangSmith tracing · evals and guardrails · API gateway with retries, circuit breakers and DLQ · Postgres + Redis · architecture diagram, eval scorecard, cost and latency numbers.
- 7+ years shipping GenAI, multi-agent RAG and vision AI systems to production
- 12+ production AI systems across 2 enterprises, 6 SMBs and 4 startups, owning the loop from scoping through deployment and handoff
- Led a $2M AI transformation for a $90B AUM PE firm: ontology-first deal evaluation that cut diligence from 4 months to ~1 hour. 6-agent ingestion pipeline over 20,000+ pages/month, production RAG-KAG on Azure, 93% Recall@10 on a 100+ question golden set
- 2x AI founder: ARTIAL (Techstars, autonomous drone navigation) and Modalina AI (NVIDIA Inception, multimodal vision AI)
- BSc in Mechatronics & Robotics; left a PhD to build ARTIAL
- Autonomous Drone Navigation: Fully autonomous visual navigation on NVIDIA Jetson Nano (4GB) in C++/ROS with ORB-SLAM3 and PX4. Live flights: indoor, outdoor
- Construction Vision AI Agent: 5 VLM agents with hybrid RAG for real-time safety analysis on GCP; saved supervisors ~15 h/week. Presented at BuiltWorlds AI Conference.
- Multi-agent RAG Platform: 5-agent platform that cut technology adoption for field managers from 6 weeks to 3. LangGraph, LangSmith, Milvus, Google App Engine.
- Agentic GraphRAG Copilot: Embedded copilot over a ~3,000-entity knowledge graph that cut cross-team data retrieval from hours to minutes, with a DeepEval regression suite.
- Enterprise MCP Server: Exposed 18 internal APIs to agents via FastMCP/FastAPI, cutting integration effort by 60%.
Client systems above were delivered under NDA, so that code stays private. The Now building repos are open rebuilds of the same patterns on public data.
Python TypeScript LangGraph LangChain LangSmith LlamaIndex MCP GraphRAG Ontology / Knowledge Graphs Neo4j Milvus Palantir AIP PyTorch FastAPI Docker AWS GCP Azure ROS C++
Contact: egor@efolley.com