Skip to content
View enached134-ctrl's full-sized avatar
🎯
Focusing
🎯
Focusing
0 followers · 1 following

Block or report enached134-ctrl

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
enached134-ctrl/README.md

Hi, I'm Daniel 👋

AI Engineer. I build production AI systems you can audit — RAG with real citations, MCP servers, and LangGraph agents, each with an evaluation suite wired into CI so quality is measured and a regression fails the build before it reaches a user. Everything here is tested, documented code you can read before you hire me.

🔗 Portfolio → enached134-ctrl.github.io

🔧 What I've shipped (open source)

  • agentic-rag-mcp — multi-agent RAG over the Model Context Protocol. Every answer cites its exact source chunk or the agent refuses; groundedness, citation and refusal evals run in CI.
  • mcp-vitals — a CLI + GitHub Action that grades any MCP server A–F for reliability and agent-usability. I graded the official reference servers — two got an A, and the sharpest finding wasn't a vulnerability; it was a tool name. Live report →
  • groundcheck — fine-tuned a 1.5B model (QLoRA) into a local groundedness judge that agreed with a frontier judge 100% of the time, at $0 per call (base F1 0.083 → 1.000).
  • shipgate — a drop-in eval gate for CI: it scores your eval cases and fails the build when quality drops below a threshold or regresses against a saved baseline. Deterministic checks run at $0 (no API key); an LLM judge is optional. Evals decide what ships.
  • AbstentionBench — an original meta-eval benchmark for the answer-or-abstain decision: 270 cases (real SQuAD 2.0 traps + an adversarially-verified hard tier) that measure whether eval tools can catch over-refusal — the failure faithfulness metrics structurally can't see. Honest leaderboard, only measured numbers. Live leaderboard →
  • reelforge — an agentic video engine on LangGraph + Claude: a supervisor orchestrates tool-calling sub-agents with deterministic gates.

🧰 Stack

Python · TypeScript · LangGraph / LangChain · Claude & OpenAI APIs · pgvector · FastAPI · Next.js · Docker · Kubernetes · GitHub Actions · PyTorch / QLoRA

The measurement decides what ships.

📩 enached134@gmail.com

Pinned Loading

  1. agentic-rag-mcp agentic-rag-mcp Public

    Multi-agent RAG (LangGraph) exposed as an MCP server: plan -> retrieve -> synthesize -> self-critique with citations. Claude + pgvector + Voyage + Firecrawl.

    Python 1

  2. reelforge reelforge Public

    Agentic short-form video engine (LangGraph + Claude): one topic -> per-scene image + video prompts, image-first and model-accurate (Nano Banana Pro / Seedance 2.0 / Kling 3.0).

    Python

  3. groundcheck groundcheck Public

    A local groundedness judge for RAG: QLoRA-distilled to match a frontier judge 100% at $0/call. Ships only if its own evals beat baseline.

    Python

  4. mcp-vitals mcp-vitals Public

    Reliability grades for MCP servers - behavioral + agent-usability evals you run yourself, in CI.

    Python

  5. shipgate shipgate Public

    A drop-in eval gate for CI — fail the build when your LLM app's evals regress. Evals decide what ships.

    Python