Siya Goel Medhya Goel Andreas Lorgen J Yim

Image chatbot prototype

This branch includes an initial image-aware chatbot app.

Run locally

1. Install dependencies: pip install -r requirements.txt
2. Set your OpenAI key: export OPENAI_API_KEY="your_key_here"
3. Start the app: streamlit run chatbot_app.py

What it does

• Upload one or more images.
• Ask natural-language questions about those uploaded images.
• Returns an answer grounded in the visible content (or says when evidence is uncertain).

---

Venn mobile app (Expo)

This repo also contains a React Native (Expo Router) client under app/. It does not replace the Streamlit prototype above; they are separate entrypoints.

Run locally

1. Use Node 20+ (see package.json engines and optional .nvmrc).
2. npm install
3. Copy .env.example to .env and set:
   • EXPO_PUBLIC_SUPABASE_URL
   • EXPO_PUBLIC_SUPABASE_ANON_KEY
4. npm start, then open in Expo Go, or press i / a / w for simulator / web.

Run on iOS simulator (i.e. to test share-sheet functionality)

0. Make sure that XCode and iOS simulators are installed on your Mac. You must also be added as an owner in the expo account associated with the build.
1. Use Node 20+ (see package.json engines and optional .nvmrc).
2. npm install
3. Copy .env.example to .env and set:
   • EXPO_PUBLIC_SUPABASE_URL
   • EXPO_PUBLIC_SUPABASE_ANON_KEY
4. In one terminal, run npx expo start.
5. In another terminal, run npx eas-cli@latest build --platform ios --profile ios-simulator and then open the simulator by running open -a Simulator.

Library search (what it does right now)

Search and light “clustering” for the Library tab (app/(tabs)/archive.tsx) live in lib/archiveSearchAndCluster.ts. Data is on-device unless teammates wire the integration hooks in lib/teamIntegrationPlaceholders.ts.

Piece	Behavior
Index	Bundled images in assets/files plus user uploads copied under the app document directory.
Query	Query string is tokenized; every token must appear somewhere in the item’s searchBlob (AND semantics), not one long substring.
searchBlob	Lowercased join of id, file name, auto-tags, theme, and optional supplemental text (for future OCR/captions).
Auto-tags	Inferred from the file name (with stopwords); stored in venn-archive-item-meta.json.
Theme chips	Coarse keyword buckets (e.g. food, travel)—not embedding-based clusters.
Ranking	Among matches, file name and tag hits score above weak blob-only hits; sort by score, then file name.
Match hints	While searching, a tile may show short labels (e.g. “File name”, “Tag”) for transparency.
Vision / OCR	Off by default. When enabled, text from placeholder_extractSearchableTextFromImage merges into the same index via lib/archiveSupplementalSearchText.ts.
Backend	lib/teamIntegrationPlaceholders.ts defines no-op hooks for Supabase tags, embedding themes, index push, chat, and vision—flip flags and implement when ready.

The Action tab is a placeholder chat UI for future generative features; it does not use chatbot_app.py.