Symposium RAG · Semantic Cache
A document-grounded symposium assistant with persistent exact and semantic answer caching. Visitors ask about a fictional event guide and receive answers with page references; repeated questions can reuse a source-preserving cached answer.
The problem
A document-grounded symposium assistant with persistent exact and semantic answer caching. Visitors ask about a fictional event guide and receive answers with page references; repeated questions can reuse a source-preserving cached answer.
My contribution
I integrated PDF retrieval, source-page references, persistent semantic caching, shared request admission, and the deployed browser interface.
Product features
- A public browser chat answers questions about a fictional, contact-free symposium PDF.
- OpenAI embeddings and FAISS retrieve guide passages for grounded answers.
- Exact and semantic cache lookups retain the original source references.
- The hosted Upstash Redis cache shares entries across serverless requests.
- Shared request admission and controlled errors bound the public demo.
Engineering work
- Source-preserving reuse: Check exact and embedding-similarity cache entries before generating a new grounded answer.
- Shared hosted state: Persist cache entries in Upstash Redis across Vercel requests; retain local adapters for development.
- Grounding & admission: Retrieve FAISS passages, keep page references, and gate client and provider requests with shared quotas.
Outcome & scope
Reuses grounded answers across repeated questions. Live checks confirmed a cache hit with source references and a refusal for absent information; production cost savings and hit rates have not been measured.
Public fictional-event demo · Persistent semantic cache · Source-page references