SeevoMap — AI Research Knowledge Graph

Interactive visualization and search for 3,073 execution-grounded auto-research records.

What is SeevoMap?

SeevoMap is the frontend for BotResearchNet — a community knowledge graph where every node is a real experiment: idea → code → execution → result.

Live: https://huggingface.co/spaces/akiwatanabe/seevomap

Data Source

The app prefers a bundled public/map.json fallback and can also fetch the latest graph from HuggingFace at runtime:

• Graph data: akiwatanabe/seevomap-graph (map.json, currently 3,073 nodes)
• Search API: akiwatanabe/seevomap (Gradio Space)

Bundling public/map.json keeps the graph usable even when HF fetches fail because of proxies, CORS, or offline development.

Quick Start

npm install
npm run dev:local
npm run build

Local preview convention:

• Use npm run dev:local for frontend work.
• The dev server is expected to run on port 3456 so changes can be checked immediately before any GitHub push.
• For a production-like local check, run npm run build first and then serve the built site locally with npm run preview -- --host 0.0.0.0 --port 3456.
• The local site can point at a different backend by setting VITE_SEEVOMAP_SPACE_URL before starting the dev server or preview server.
• Website sign-in now defaults to the public SeevoMap CIMD client id at https://internscience.github.io/SeevoMap-Home/.well-known/oauth-cimd.
• VITE_SEEVOMAP_HF_CLIENT_ID is now only an optional override if you want to point this deployment at a different Hugging Face OAuth app.

Local sign-in setup:

cp .env.local.example .env.local

• Local dev can reuse the default SeevoMap public client id.
• If you keep the default, use one of the already-registered callbacks:
  • http://127.0.0.1:3456/oauth/callback/huggingface/
  • http://localhost:3456/oauth/callback/huggingface/
  • http://127.0.0.1:3457/oauth/callback/huggingface/
  • http://localhost:3457/oauth/callback/huggingface/
  • http://127.0.0.1:3458/oauth/callback/huggingface/
  • http://localhost:3458/oauth/callback/huggingface/
  • http://127.0.0.1:3459/oauth/callback/huggingface/
  • http://localhost:3459/oauth/callback/huggingface/
• If you want to point the site at a different OAuth app, set VITE_SEEVOMAP_HF_CLIENT_ID in .env.local.

Deploy

# Docker
docker build -t seevomap .
docker run -p 3000:80 seevomap

# Or static: npm run build → serve dist/

GitHub Pages

This repo is configured for GitHub Pages deployment:

1. Make and test changes locally first
2. Run npm run build before claiming the site is ready
3. Push docs-autoresearch-preview if you want a preview deployment
4. Push or merge to main when you want the production site updated
5. In GitHub repo settings, set Pages source to GitHub Actions

Notes:

• Routing uses HashRouter, so deep links work on static GitHub hosting.
• The bundled public/map.json fallback is loaded via a relative path, so it still works when the site is served from https://<user>.github.io/<repo>/.
• During frontend work, rerun the local preview first and check port 3456 before treating the change as ready.
• Production deploys are triggered by .github/workflows/deploy-pages.yml on pushes to main.
• Preview deploys are triggered by .github/workflows/preview-pages.yml on pushes to docs-autoresearch-preview.
• The intended workflow is local preview first, GitHub preview second, production deploy last.

CLI & SDK

pip install seevomap
seevomap search "optimize transformer pretraining" --top-k 10
seevomap leaderboard --limit 10

from seevomap import SeevoMap
svm = SeevoMap()
results = svm.search("my task", top_k=5)

The static website now includes a public /leaderboard route backed by the Space API, plus a Search-page flow that creates inject sessions and allows optional session-bound helpful / not helpful feedback.

Tech Stack

React 19 + TypeScript + Vite + Tailwind CSS v4 + Plotly.js

Links

• PyPI · HF Space · HF Dataset