About
Applied AI, with judgment
My edge is not training models: it is specifying, building and evaluating GenAI systems that work in production. I come from Requirements Engineering, so every agent is born with use cases, business rules, guardrails and evaluation metrics.
I evaluate LLMs on international platforms (Turing, Outlier AI) with SxS and fact-checking, and my MSc research (UFPB) evaluates the quality of user stories generated by different LLMs, with blind analyst review.
Trajectory
AI Experience
Applied GenAI in a regulated environment
Practical use of LLMs inside the requirements-engineering flow of critical public-health systems.
- Accelerated regulatory analysis and test-scenario generation with GenAI.
- Consistency validation across business rules, screens and integrations.
LLM evaluation and alignment
International model-quality work: SxS, fact-checking and error documentation for RLHF.
- Side-by-Side scoring for prompt adherence, completeness and consistency.
- Contributed to hallucination reduction and model alignment.
Conversational agents in production
Building and operating real agents: education chatbots, a WhatsApp scheduling SaaS and the multi-agent platform behind this site.
- Typebot + n8n + LLM APIs (ChatGPT, Gemini) with guardrails and handoff.
- Own platform: FastAPI + Postgres + ChromaDB + Gemini, multi-tenant, running on this VPS.
Hands-on
AI projects
AWS case: document analysis with GenAI
Documented portfolio project
RAG platform for regulated environments with Bedrock Knowledge Bases (Titan V2 + Nova Lite), S3 Vectors and Lambda. Synthetic corpus with ground-truth validation: 3/3 correct answers with citations. Bugs found and documented.
GitHub →This site's RAG + guardrails chatbot
In production now
Layered lexical retrieval, conversation memory, anti-injection and handoff. Transparent pipeline: every answer shows guardrails, chunks and latency. A deliberate choice not to use embeddings on a small base.
GitHub →LLM evaluation and alignment
Turing · Outlier AI
Side-by-Side evaluations, fact-checking with evidence validation and error-pattern documentation for RLHF, under international quality standards.
GitHub →Education chatbots
LIFEE
Conversational agents with Typebot + n8n + ChatGPT API: personas, flows, rules and iterative tuning, with fewer out-of-context replies.
GitHub →Multi-tenant SaaS platform with AI agents
Nil Alisson AI Solutions
Platform where businesses create and publish their own AI agent: site with a visual editor, WhatsApp with text/audio/image, admin panel and custom domain. Evolution API, PostgreSQL, FastAPI and PHP.
GitHub →ReqGuard: RAG for Requirements Engineering
Applied research · in progress
RAG system with LangChain/LangGraph, ChromaDB and RAGAS evaluation over a synthetic regulatory corpus, exploring how AI agents can support requirements engineering in critical environments.
GitHub →RAG provider benchmark
Vertex AI vs. Gemini API
Comparison of two RAG architectures using RAGAS metrics, evaluating response quality and source citation between the two providers.
GitHub →Research: do LLMs write good requirements?
MSc UFPB
Experiment comparing user stories generated by DeepSeek-R1, GPT-4o, Sabiá-3, LLaMA and Gemini, with blind analyst evaluation and requirements-quality criteria.
GitHub →Tooling
Working stack
GenAI & LLMs
AWS
Automation & Agents
Foundation & Data
In production
AI Agents
Specialized agents you can test right now, each with RAG, guardrails and human handoff.
Contact
Let's build with AI
Open to remote roles in applied GenAI, AI functional analysis and a path into agent engineering (MCP, RAG, integrations).
Validation
Certifications
GenAI for Startups (AWS Bedrock)
Sebrae
AWS Certified Cloud Practitioner
Amazon Web Services
Scrum Foundation Professional (SFPC)
Certiprof
Agile Methodologies: Lean, Scrum & Squads
DIO
Trajectory
Education
MSc in Computer Science
UFPB · 2024–2026 · Research: LLMs applied to Requirements Engineering