Project Overview
Objective
Built room-scoped semantic search and RAG question answering over multilingual chat history, images, and OCR-extracted content.
Stack
FastAPINext.jsPostgreSQLQdrantSentenceTransformerCLIPBLIPGoogleTranslatorTyphoon OCROpenAI
Delivery highlights
- Developed a proof-of-concept chat application where users create or join rooms, send text and image messages, search conversation history, and ask natural-language questions grounded in retrieved messages. Stored structured room and message data in PostgreSQL and indexed multilingual text, CLIP image embeddings, BLIP captions, and Typhoon OCR output in Qdrant. Combined semantic and keyword retrieval with an OpenAI LLM to answer questions within one room, across rooms, or about shared images through a FastAPI backend and Next.js chat interface.
Back to Topic ProjectsBack to All ProjectsSystem workflow
How chat history becomes searchable knowledge
Messages and images are enriched, indexed, retrieved by room context, and used to generate grounded answers.Rendering workflow diagram…
Scroll horizontally to explore the full workflow on smaller screens.
System workflow
Multimodal message indexing pipeline
Text, images, and OCR output are enriched and embedded into Qdrant for unified semantic search across all message types.Rendering workflow diagram…
Scroll horizontally to explore the full workflow on smaller screens.
System workflow
Room-scoped hybrid retrieval
Questions are matched using both semantic vectors and keyword search, filtered by room membership, and answered with LLM grounding.Rendering workflow diagram…
Scroll horizontally to explore the full workflow on smaller screens.
System workflow
Source-reviewed processing sequence
A conservative end-to-end boundary taken from the source audit. Specific endpoints, storage writes, and external calls are only claimed in the project-specific diagrams when the implementation confirms them.Rendering workflow diagram…
Scroll horizontally to explore the full workflow on smaller screens.