MEDIA PREVIEW
Overview & Result
Developer Aayan built YC Indexor, indexing all 6,000+ YC portfolio companies across every batch. By leveraging Jev instead of traditional LLM embeddings or brute-force GPT queries, indexing 90,000,000 tokens of founder bios, company one-liners, and technical tags cost just $2.70 in total. Users can search queries like 'hard-tech hardware startups founded by ex-SpaceX engineers' and receive instant results.
How Jev fits in the loop
- Scraped public Y Combinator directory data for over 6,000 companies across all batches.
- Batch-processed company profiles through Jev to extract standardized industry taxonomy and tag probabilities.
- At search time, runs real-time Jev relevance scoring against candidate pools to rank matching companies in under 150ms.
- Serves a fast web interface with live filtering, tag clouds, and verified portfolio insights.
How to reproduce
- Clone aayan/yc-indexor and run npm install.
- Check data/yc_companies.json containing scraped YC company profiles.
- Run npm run index to stream records through Jev Choice classifier.
- Launch web UI with npm run dev and query arbitrary semantic tags like 'deeptech fusion ex-nasa'.
Why this build matters
Shows the immense cost and throughput advantage of Jev for knowledge indexing: $2.70 for 90 million tokens is orders of magnitude cheaper than OpenAI batch embeddings or Claude text-processing.
Reported performance
Reported by authorCost: $2.70 / 90M tokens
Latency: 120ms / query
Indexed 6,000+ startup bios (90M tokens total) for $2.70. Real-time search executes in ~120ms without maintaining a vector DB.
Limitations
- Covers publicly available YC directory text; private financial metrics or unlisted founders are excluded.
- Best suited for semantic tag filtering rather than full natural language conversational chit-chat.