I showcase six AI‑powered developer tools that I built as LangGraph agents with strong safety and security guardrails. Each project includes a short tagline, core capabilities, quantitative results, and the tech stack I used.
- DeployMate AI – 7‑agent LangGraph copilot that turns a brief + OpenAPI spec into requirements, field mapping, architecture, test plan, risk register and a handoff doc in one run.
DeployMate AI was evaluated on 40 cases and achieved 100% architecture validity.
OpsPilot AI obtained 98% retrieval Recall@5, 99.6% groundedness, and 84% task completion on 50 scenarios.
Toolgate covers 10 threat scenarios mapped to named tests.
RAG‑Guard’s synthetic drift benchmark improved recall from 71.7% to 100% and reduced hallucination from 1.7% to 0%.
PR Sentinel enforces 12 security rules plus LLM review.
Postpilot was validated with 46 offline tests.
- OpsPilot AI – Incident‑investigation agent with RAG over 20 runbooks, JWT RBAC, SQL allowlisting, PII masking and admin approval before any change.
- Toolgate – MCP data‑governance agent that enforces role‑based column grants, row filters and masks server‑side.
- RAG‑Guard – Retrieval‑quality gate that detects degraded retrieval and auto‑escalates through BM25, hybrid RRF and keyword fallbacks.
- PR Sentinel – Secure CI‑compatible code‑review automation with HMAC‑verified webhooks and idempotent processing.
- Postpilot – Human‑approved social agent that orchestrates research‑to‑publish flow with checkpointed approvals.