Frédéric Masi · London, UK

I turn hard technology into adoption.

Technical GTM and partner-ecosystem leader across telco, cloud-native, and open source. On the side, I build private, self-hosted AI agents — where your data stays on your own hardware.

15+ years across Red Hat, Ericsson, Canonical, and Wind River — partner ecosystems, solution-engineering enablement, and developer- and partner-facing programmes that drive real adoption of complex platforms (Kubernetes, OpenStack, telco cloud, 5G/O-RAN). I built and ran Wind River's Ready for Wind River partner certification and validation programme. Lately I've been building hands-on with AI: retrieval, on-device models, and agentic tooling.

Things I've built

mailrag

Private, self-hosted email RAG. Turn your own mail archive into a queryable knowledge base on your hardware — hybrid retrieval, thread-aware answers, optional local-LLM cleanup.

Python · LlamaIndex · Qdrant · open source

Parley

Private, fully on-device meeting transcription for Apple Silicon. Dual-stream mic and system capture, speaker diarization, echo removal, crash-safe — zero cloud.

Swift · on-device ML · open models · open source

cycling-analysis-agent

An AI cycling coach you host yourself. Reads your rides, predicts your routes, tracks your form, and verifies every climb against 1-metre LIDAR. A Claude agent with a directeur sportif's attitude.

Python · agentic AI · self-hosted · open source

Writing

Two techniques doubled my email search — and neither was a fancier model

I rebuilt email search from scratch and measured every technique on 360 real questions. Recall went from 46% to 93% — and the two biggest wins weren't a fancier model.

retrieval · learned-sparse hybrid · benchmarks