diogo.brancoget in touch →
building agentic AI·Lisbon, Portugal·UTC+00 / WET

Diogo Branco

AI Engineer/Agentic AI & RAG Systems

I build agentic AI systems — from multi-agent to agentic RAG — and shipped one into production at Portugal's largest telco.

ask‑my‑cv
preparing

An agent answers from Diogo's thesis & CV, showing its tool calls and citations. It's an LLM — it can be wrong.

90%+
RAG answer accuracy
up from ~40%
50k
indexed chunks
3,000+ documents
<30s
query latency
per frontline query
01/ about

I'm an AI Engineer who shipped an agentic RAG system to the call-centre and retail operations of Portugal's largest telco, lifting answer accuracy from ~40% to over 90% for thousands of frontline staff. On the side I build agentic systems of my own — a legaltech platform for Portuguese law firms and an agentic copywriting pipeline for a social-media agency — and my MSc thesis fine-tuned Mistral-7B for clinical-trial natural-language inference, benchmarked against the other SemEval-2024 systems.

AI is reshaping every industry — for good and for bad. I want to be one of the people who build it for good.

Off the clock, you'll usually find me at the gym, deep in a video game, or out for a drink with friends.

Education

  • MSc Computer Science2021 — 2024
    Instituto Superior Técnico

    Specialisations in Artificial Intelligence & Distributed Systems. Thesis: "Adapting Large Language Models for Clinical Trials".

  • BSc Computer Science2018 — 2022
    Instituto Superior Técnico
  • Blockchain & Smart Contracts (Advanced Course)2022
    IST & Universidade Católica Portuguesa

Languages

Portuguese (native) · English (fluent)

02/ work

AI Engineer

Sep 2025 — Present
MEO (Altice Portugal) · Centro de Inteligência Artificial
  • Spearheaded development and production deployment of an enterprise RAG system for internal knowledge management, raising answer accuracy from ~40% to over 90% and advancing to technical leadership on the project.
  • System serves the call-centre and retail operations of Portugal's largest telco — indexing 3,000+ heterogeneous documents into ~50,000 chunks and answering queries in under 30 seconds.
  • Member of the Agentic squad responsible for the company's agentic AI solutions, contributing to architecture and tooling decisions alongside the tech lead.
  • Builds production agentic solutions with Microsoft Agent Framework, Semantic Kernel and LangChain, integrating external tools and data via Model Context Protocol (MCP).
  • Owns RAG releases end-to-end: Docker, Azure Container Apps via CI/CD, A/B tests and RAGAS-based evaluation before rollout.
  • Contributed to the cloud architecture of MEO's AI Factory and built data ingestion pipelines feeding Microsoft Fabric (Lakehouse).

Junior Software Engineer

Feb 2023 — Jun 2025
DataMorfose
  • Automated internal company workflows to optimise operational productivity.
  • Conducted advanced data analytics and built automation pipelines to improve data insights.
03/ projects

AI Legal Team Platform

building
2025 — Present

A legaltech platform (internally "GaiusLaw") for Portuguese law firms — document drafting, legal research, case management and court-notification triage, entirely in European Portuguese.

  • Ingests court notifications from Gmail and the Citius platform, auto-classifies them, matches them to the right case, computes Portuguese filing deadlines, and drafts replies.
  • Hybrid-search RAG (pgvector + BM25 fused with RRF, plus a cross-encoder reranker) over Portuguese legislation and court decisions, with drafting and research agent pipelines.
RAGHybrid SearchLegalTechFastAPIpgvector

Agentic Copywriting Automation

building
2026 — Present

An agent that automates the copywriter role in a short-form video pipeline (TikTok / Reels / Shorts), turning a human ideation into a production-ready script.

  • Turns a human-produced ideation into a formatted script with captions, hashtags and visual directions, written in the client's brand voice.
  • A staged pipeline (parse ideation → load references → fetch resources → draft via Claude → validate → export a color-coded PDF), with a human approving every script before it ships.
Agentic AIClaude APIPipelineHuman-in-the-loop

Clinical Trials NLI System

thesis
2024 — 2025

Natural language inference over clinical trial reports — the core of his MSc thesis and a SemEval-2024 NLI4CT submission.

  • Fine-tuned Mistral-7B on the clinical-trials domain, combined with a chain-of-thought few-shot approach for robustness on complex medical text.
  • Built GPT-4o data-augmentation pipelines and achieved competitive SemEval-2024 NLI4CT results with strong faithfulness metrics.
Fine-TuningMistral-7BChain-of-ThoughtSemEval

More in the works

A few other agentic systems in progress. Ask the assistant about them, or get in touch.

04/ stack

AI & ML

  • LLMs
  • RAG
  • Agentic AI
  • Multi-Agent Systems
  • Microsoft Agent Framework
  • Semantic Kernel
  • LangChain
  • LangGraph
  • MCP
  • Fine-Tuning
  • Prompt Engineering
  • Embeddings

Eval & MLOps

  • RAGAS
  • LLM-as-Judge
  • A/B Testing
  • CI/CD
  • Docker
  • Azure Container Apps
  • Model Observability

Cloud & Data

  • Azure AI Foundry
  • Azure OpenAI
  • Azure AI Search
  • Container Apps
  • Blob Storage
  • Functions
  • Microsoft Fabric
  • Data Pipelines

Languages

  • Python
  • TypeScript
  • JavaScript
  • SQL

Web

  • Next.js
  • React
  • FastAPI
  • Node.js

Tools

  • Git
  • GitHub Actions
  • Claude Code
  • Agile / SCRUM
05/ contacts

Let's build something.

Based in Lisbon, Portugal, building agentic AI. Always happy to talk shop — agents, RAG, applied LLMs. Drop me a line.

Download CV (PDF) ↓