Agentic lead-research workspace that compiles natural-language ICPs into executable company and people research workflows, combining structured providers, web research, and row-level enrichment with traceable evidence.

Architected a secure code-execution plane where LLM-generated TypeScript is parsed, type-checked, AST-policy validated, source-hashed, and bound to an approved capability manifest before running inside a fresh permission-restricted Deno worker.

Built bounded JSONL RPC between the isolated worker, a Node.js supervisor, and an authenticated Python gateway using opaque workspace handles; added SSE progress streaming, PostgreSQL-backed tables, cursor-aware lead sources, durable execution artifacts, cancellation, resource limits, and scheduled Celery automations.

Multi-tenant SaaS platform for AI-powered GTM automation across Reddit, X, YouTube, and HackerNews.

Engineered a competitive intelligence engine using crawler agents and RAG pipelines for 10x faster market analysis.

Architected a Dockerized multi-agent orchestration backend with scheduled execution, real-time SSE streaming, and credit-based billing.

Autonomous, AI-powered system administrator built on LangGraph with persistent memory and sequential reasoning.

Implemented a multi-agent architecture (Planner → Executor → Reviewer) with human-in-the-loop SSH execution.

Exposed 60+ infrastructure tools via a FastMCP server supporting SSE/stdio transports for Claude Desktop integration.

Knowledge-sharing platform featuring an intelligent AI Assistant built to answer company-specific queries instantly.

Engineered a powerful RAG pipeline using Qdrant vector embeddings to retrieve context directly from shared internal documents.

Integrated an advanced LLM backend for highly accurate, context-aware summarization and conversational generation.