Software Engineer · AI Systems

I build AI systems that listen, reason, and coordinate.

My work sits around the model: real-time voice pipelines, context engineering, distributed services, and protocols that let intelligent agents work together.

Latency
Fast enough to feel natural
Grounding
Context the model can trust
Reliability
Failures designed, not discovered
Observability
Behaviour you can explain
01

Talkdesk · Real-time voice AI

A conversation is a distributed system with no patience.

I work on backend systems for enterprise AI, including real-time voice experiences where audio has to become understanding—and understanding has to become speech—without breaking the flow of a conversation.

The interesting engineering lives between the models: streaming, turn detection, interruption handling, state, latency budgets and graceful recovery.

  • Streaming audio
  • VAD
  • STT
  • Agent runtime
  • TTS
  • Barge-in
  • LiveKit
Realtime voice loop streaming
01Audiolive stream
02VADspeech events
03STTpartial text
04Agentreason + act
05TTSstreamed voice
barge-ininterrupt → cancel → resume
Latency-sensitive orchestration across speech, reasoning and response.
02

Scooli · Context engineering

Better output starts before the prompt.

At Scooli, we turn curriculum, lesson objectives and teacher context into grounded classroom workflows. The goal is not generic generation—it is material that fits what a teacher needs to teach next.

That means designing retrieval, ranking, context assembly and structured generation as one system, with the teacher staying in control of the final result.

  • Context engineering
  • RAG
  • pgvector
  • PostgreSQL
  • Quarkus
  • Structured output
Visit Scooli
Context enginegrounded generation
Curriculum
Teacher context
Lesson objective
Context assemblyretrieve · rank · compose
Structured generationteacher-ready material
Retrieval is only one step; useful output depends on assembling the right context.
03

MSc research · Agent interoperability

Useful agents need a shared way to work together.

My thesis explores agent-to-agent interoperability in enterprise systems: how independently built agents discover capabilities, authenticate, exchange tasks and coordinate across system boundaries.

The proof of concept treats A2A as infrastructure rather than a demo—explicit contracts, observable task states and predictable failure handling.

  • Google A2A
  • Agent discovery
  • JSON-RPC
  • SSE
  • Authentication
  • Task lifecycle
Agent networkA2A task lifecycle
Agent Acapabilities
Agent Btask owner
Agent Cspecialist
A2A gatewaydiscover · authenticate · delegate
task: working
Independent agents coordinating through explicit capabilities and task states.
Pedro Rocha outdoors.

About

The system around the intelligence is what makes it real.

I’m a software engineer in Talkdesk’s AI Engineering team, an MSc researcher at ISEP, and a co-founder of Scooli. My background is in backend engineering; my focus is building AI products that remain understandable, observable and reliable after the first impressive demo.

TalkdeskScooliMSc AI EngineeringPorto, Portugal

Contact

Let’s talk about AI systems—or a problem that needs one.