Skip to content

OSA Platform Overview

An overview of the Open Science Assistant platform: architecture, community onboarding, knowledge pipeline, and roadmap.

Recording

Interactive Slides

Use arrow keys to navigate. Press F for fullscreen, ? for shortcuts.

Summary

The Open Science Assistant is an extensible AI platform that serves small research communities from a single lab server. Key points covered:

  • Design philosophy: Precision over features; researchers need accurate, citation-backed answers
  • Architecture: Cloudflare Worker edge proxy, FastAPI + LangGraph backend, SQLite FTS5 knowledge stores
  • Community onboarding: One YAML file creates a full AI assistant with API routes, tools, and knowledge sync
  • 6 live communities: HED, EEGLAB, BIDS, MNE-Python, NEMAR, FieldTrip
  • Two-tier tool system: Auto-generated knowledge tools from YAML plus custom Python plugins
  • Knowledge pipeline: 5 sources (GitHub, OpenALEX, mailing lists, Discourse, docstrings) synced into per-community databases
  • Smart FAQ generation: Two-stage LLM pipeline reducing costs by ~85%
  • Embeddable widget: One script tag to add an AI assistant to any project website
  • Future directions: Multi-assistant delegation across communities, ephemeral preview backends for self-service onboarding

Interactive slides built with Agentic Presentation Builder, a JSON-to-Reveal.js presentation tool.