What the project was for
Hyperledger's technical knowledge is spread across documentation, issues and pull requests for a dozen separate projects. The brief was a unified knowledge layer that could answer accurately across all of it, beginning with Besu and Fabric.
As one of the early contributors I helped build the production system from the ground up, rather than extending something that already worked.
The multi-agent RAG platform
- Problem
- Answers had to be grounded in the real documentation of many separate projects, and stay correct as those projects changed.
- Approach
- Designed the end-to-end RAG pipeline: ingestion from docs and GitHub issues and PRs, chunking and embedding, ChromaDB for vector search, LangGraph orchestrating the agents, and FastAPI serving it. Open-source models from Hugging Face were deployed on AWS, with Snowflake as the data platform.
- Result
- A production-ready system serving an entire open source ecosystem rather than a single application.
What it taught me
Document ingestion, embedding pipelines, vector databases, retrieval optimisation, multi-agent orchestration, cloud deployment and serving language models where other people depend on the answers.
It was my first system built for an ecosystem rather than an application, which changes what counts as finished.
The people behind it
Supported by LF Decentralized Trust and Founder Institute, and mentored by Barbara (Bobbi) Muscara, a Technical Steering Committee member at LFDT, and Gianluca Capuzzi.
Their reviews shaped how I build production AI systems, and how I think about maintainability long after a mentorship ends.
Open Source Summit Korea
The mentorship led to an invitation to speak at Open Source Summit Korea — my first international open source conference — presenting the project and the work happening inside LFDT.
It was also where I met Daniela Barbosa, Executive Director of LF Decentralized Trust, and Julian Gordon, its Vice President.
- Accepted to LFX MentorshipMulti-Agent RAG for AIFAQ
- AIFAQ presented at an LFDT meetup
- Mentorship completed