Inspiration We've all fallen into the rabbit hole — hours of YouTube, blog posts, and Reddit threads trying to figure out how the best people in a field actually got there. The information exists, but it's scattered, shallow, and impossible to act on. We wanted a system that could do that research autonomously and hand you back something you could immediately execute on. That idea became ApexIntel.

What it does ApexIntel is an autonomous multi-agent pipeline that takes a single input — a field like "venture capital" or "competitive programming" — and returns a structured Obsidian knowledge vault containing the distilled strategies of the top performers in that field. Each performer gets three notes: a strategy overview, a phased actionable task list with checkboxes, and an annotated sources file. The entire process runs end-to-end without human intervention, streamed live to a terminal-style UI.

How we built it We built ApexIntel on Google ADK with Gemini 2.5 Flash as the reasoning engine across five specialised agents. A Querying Agent generates and self-scores Tavily search queries, retrying with feedback when results fall below a quality threshold. A Tavily Extractor pulls full article content from every discovered URL. A Synthesizer Agent — three writer agents, one validator, and one repair agent — produces and quality-checks every Obsidian note. The backend is FastAPI on Google Cloud Run with Server-Sent Events for live log streaming and Google Cloud Storage for vault persistence. The React frontend provides a real-time log viewer, a file-tree notes browser, and direct links to Arize Phoenix for LLM trace observability.

Accomplishments that we're proud of We're proud of the self-healing note quality loop — the validator and repair agents working together means the output is structurally consistent regardless of what the source material looks like.

Built With

  • arize-phoenix
  • artifact-registry
  • cloud-build
  • cloud-run
  • cloud-storage
  • fastapi
  • firebase-hosting
  • gemini-2.5-flash
  • google-adk
  • javascript
  • opentelemetry
  • python
  • react
  • secret-manager
  • tavily-search-&-extract
  • vertex-ai
  • vite
Share this project:

Updates

Submission history