Skip to content
View LucasSantana-Dev's full-sized avatar
🧐
🧐

Organizations

@Forge-Space

Block or report LucasSantana-Dev

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.

Content in all repositories owned by your account will be closed.
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
LucasSantana-Dev/README.md

Hi, I'm Lucas (Luk)

Senior Software Engineer · Backend-focused Full-Stack · AI agent tooling and evals · Node.js · TypeScript · Python

Currently at Thoughtworks (remote from Brazil, UTC-3) · 5+ years shipping cloud-native systems for international clients.

Portfolio LinkedIn Email hitgate on PyPI English C1


What I do

I build backend-focused full-stack systems for a living and tools for AI coding agents on the side. I care about measuring those tools: retrieval quality in hitgate, and which tools an agent actually gets in decision-critic.

Day job is senior consulting at Thoughtworks (previously CI&T), on international teams with strict review cultures. Off-hours I run a homelab and a self-hosted Discord bot, and I publish developer tooling on npm and PyPI.

  • decision-critic: a runtime probe showed a Claude Code denylist left 91 MCP tools in place, including one that runs shell commands. The empty allow-list resolved to 0.
  • hitgate: hybrid retrieval reaches Hit@5 0.99 on a 99-case benchmark (BM25 alone: 0.91), with a label-free evaluation gate in CI.
  • Backend for a global consumer brand's mobile app and personalization platform (Node.js, AWS Lambda).
  • Lucky: staging and production deploys, a migration gate and mutation testing.

Featured projects

Project What it does Stack
decision-critic A reviewer agent with no tools, on purpose. Portable spec, Claude Code adapter, and a probe that checks what the host actually resolves Node.js · Claude Code License
hitgate Hybrid RAG evaluation harness: BM25 + dense + RRF retrieval, label-free eval, CI-ready Python · Sentence Transformers · BM25 PyPI
Lucky Self-hosted Discord music bot: Spotify stats, autoplay recommendations, React dashboard, moderation suite TypeScript · Prisma · Discord.js CI
homelab Self-hosted infra: Docker Compose, Caddy, Tailscale, Prometheus/Grafana, SOPS Python · Bash · Docker CI
sharekit Share your AI coding setup (CLAUDE.md, skills, cursorrules) with one command: npx @lucassantana/sharekit install lucassantana Node.js · CLI npm

Additional private repos (finance tools, infra dashboards) available on request.


Tech stack

Backend and cloud: Node.js · TypeScript · Python · Express · AWS (Lambda, S3, API Gateway, SSM, CloudWatch, EventBridge) · PostgreSQL · Redis · Prisma · Docker · Cloudflare Workers and Pages · Serverless Framework

AI agents: Claude Code (subagents, hooks, skills) · MCP servers · RAG retrieval and evaluation

Frontend: React · Next.js · Tailwind CSS · Material UI

DevOps and observability: GitHub Actions · Jenkins · Jest · Mocha · Grafana · Splunk · Kubernetes (k3s) · SOPS/age · Tailscale · Caddy · Linux


Certifications and education

  • AWS Certified Cloud Practitioner
  • Bachelor's in Computer Systems, FIAP (2025)

Open to

  • Open-source collaboration on AI agent tooling, evals, or observability
  • Architecture and system-design discussions. Happy to review or pair

Get in touch: lucas.diassantana@gmail.com · LinkedIn


GitHub stats

GitHub stats Top languages

Pinned Loading

  1. decision-critic decision-critic Public

    A reviewer agent with no tools, on purpose. Portable spec + Claude Code adapter, with a runtime probe that proves the empty allow-list resolves to zero tools.

    JavaScript

  2. Lucky Lucky Public

    Self-hosted Discord music bot — Spotify + YouTube + SoundCloud, autoplay recommendations, React dashboard, moderation suite. TypeScript monorepo.

    TypeScript 20 9

  3. hitgate hitgate Public

    Portable hybrid (BM25 + dense + RRF) retrieval engine and a label-free evaluation harness — extracted from a personal AI-assistant memory index and decoupled to run on any source tree.

    Python 5

  4. sharekit sharekit Public

    Share your AI coding setup — CLAUDE.md, skills, cursorrules — with one command. install / preview / rollback.

    Shell 1

  5. repaint repaint Public

    repaint — a self-contained npx-skills / Claude Code skill for frontend UI that doesn't look AI-made: research, tokens, a reference library, and a live browser slop-audit loop.

    Python 1

  6. homelab homelab Public

    Secure homelab infrastructure with monitoring, container management, and Discord notifications

    Python 1