Skip to content

MASTER.md

Latest commit

 

History

History
130 lines (90 loc) · 6.66 KB

MASTER.md

File metadata and controls

130 lines (90 loc) · 6.66 KB
Raw

ProjectBrain — Master Document

Single source of truth for Devpost entry, README, demo script, and submission. Everything derives from here.

1. Devpost Entry

Title

ProjectBrain: Persistent Organizational Memory for AI Development

Tagline

Stop AI amnesia. ProjectBrain turns your architectural decisions into a live, searchable knowledge graph that AI tools and CI/CD both respect.

Description

ProjectBrain is a persistent memory layer that cures "Context Rot" — the phenomenon where AI assistants, teammates, and CI/CD pipelines drift dangerously far from the project's architectural truth.

Built on Cognee's graph-vector hybrid store, ProjectBrain:

  • Remembers every architectural decision, incident, and metric as nodes in a knowledge graph, linked by relationships (SUPERSEDES, IMPACTS, CAUSES)
  • Searches with 17 different modes (graph completion, keyword, similarity, chain-of-thought, and more) so you can query your organizational memory the way you think
  • Enforces knowledge through a CI/CD reviewer agent that rejects PRs reintroducing superseded patterns
  • Visualizes the entire memory as a live force-directed graph, updated in real-time via SSE
  • Sessions let you experiment with context without polluting permanent memory, then promote what matters

Architecture: Three lightweight processes — an SSE-backed REST API (FastAPI, port 8000), an MCP server for IDE integration (Cursor, Claude Code, any MCP host), and a Next.js 16 dashboard — all wired to Cognee Cloud.

The result: Your AI tools never forget what your team decided last week. Your CI/CD never approves a PR that reintroduces a pattern you deprecated last month. Your dashboard shows you the living history of your project's architecture.

Tech Used

  • Cognee Cloud — graph-vector knowledge graph, 17 search modes, add/cognify/search/improve/forget verbs
  • FastAPI — SSE streaming backend
  • FastMCP — Model Context Protocol (MCP) server for IDE tool integration
  • Next.js 16 — React 19, App Router, force-directed graph dashboard
  • react-force-graph-2d — D3 force simulation visualization
  • uvicorn — ASGI server
  • Tailwind CSS 4 — Dashboard styling
  • Groq (llama-3.3-70b-versatile) — LLM for graph completion search
  • GitHub Actions — CI/CD reviewer agent

Collaborators

Solo entry.

Category Tags

Developer Tools, AI/ML, Open Source, Infrastructure

2. README Content

(Blocks ready for README.md. Current README.md should be overwritten with this content before submission.)

See the existing README.md — it's already strong. At submission time, update it from this section for freshness.

Key changes to make in README before submission

  • Update Architecture diagram to show Cognee Cloud (not local Kuzu/LanceDB)
  • Add MASTER.md reference in Demo section
  • Update judging criteria table if new features (graph makeover, meta-narrative, safety net) are implemented

3. Demo Script

Total target: 3:00–4:00 minutes Tool: OBS Studio or Xbox Game Bar

Time Visual Audio
0:00–0:15 Title card: "ProjectBrain — Persistent Organizational Memory" Background music fades in
0:15–0:45 Dashboard opens at Railway URL. Seed data graph visible. "This is ProjectBrain. It's a persistent memory layer for your codebase. Every architectural decision, every incident, every metric — it's all here as nodes in a knowledge graph. This graph updates in real-time."
0:45–1:15 Click a node → detail panel slides in. Show SUPERSEDES edge. "Click any node to see its details. Notice this decision to use Postgres — it supersedes the earlier MongoDB decision. The graph captures both the decision and why it changed."
1:15–1:45 Switch search mode dropdown. Type a query. Show results. "Search across modes — graph completion, keyword, similarity, chain-of-thought — all 17 Cognee search modes are available. This is how you query your organizational memory."
1:45–2:15 Switch to terminal. Run demo_seed.py. Graph grows live. "Now let's see the memory grow. I'm adding new decisions one at a time. Watch the graph update with each entry — the SSE stream pushes updates to the dashboard in real-time."
2:15–2:45 Show PR with the reviewer_agent CI check (if ready) or show MCP tool via Cursor. "ProjectBrain also enforces memory. The CI/CD reviewer agent checks every PR against the knowledge graph. Try reintroducing a deprecated pattern — it gets blocked. Or use the MCP tools directly from your IDE to remember decisions without leaving your editor."
2:45–3:00 Scroll metrics bar. Show node/edge counts, latency, memory composition. "Real-time metrics show you graph health — node and edge counts, search latency, memory composition breakdown. Everything is live."
3:00–3:15 Closing card: GitHub repo + Cognee logo "ProjectBrain: because your team's decisions should be the last thing your AI forgets. Built on Cognee for the WeMakeDevs × Cognee Hackathon."

Pro tips

  • Pre-warm: start api.py, seed.py, and dashboard before recording
  • If using Railway URL, make sure it resolves before the screen recording starts
  • Use a large font in the terminal for demo_seed.py
  • Record in 1080p, 30fps
  • Add subtitles (YouTube auto-captions are fine)

4. Submission Checklist

Before recording the demo video

  • api.py starts without error
  • seed.py runs and populates graph
  • Dashboard renders at Railway URL (or localhost:3000)
  • SSE stream works (graph is live)
  • Search returns results for ~2 search modes
  • Node click panel works (if D3 implemented)
  • Edge filter works (if D3 implemented)
  • demo_seed.py runs and graph grows incrementally (if D5 implemented)
  • Metrics bar shows real data
  • reviewer_agent.py passes (or skip if not needed for demo)

Before Devpost submission

  • Demo video recorded, rendered, uploaded to YouTube (unlisted) — URL saved
  • README.md updated from MASTER.md section 2
  • Devpost entry filled from MASTER.md section 1
  • GitHub repo is public (or accessible to judges)
  • Railway deployment is live and URL tested
  • Tags selected: Developer Tools, AI/ML, Open Source, Infrastructure
  • Collaborators confirmed
  • "Built with Cognee" badge in README

Final sanity checks

  • Open incognito window, navigate to Railway URL — everything works
  • Demo video plays through without glitches
  • README renders on GitHub without broken links
  • Devpost preview looks clean
  • Submission submitted before Jul 5 deadline

Generated 2026-07-02. Update before each recording/submission pass.