AI Research Assistant — Quiet Links
Case Study Summary
Client: Tim Gallati, Founder of Quiet Links
Website: quietlinks.com
Industry: Scientific Research & Knowledge Management
Vertical: AI for Technical Documentation
Results:
- Complete RAG system delivered in 6 weeks
- Saved client an estimated 1 year of solo development
- 200+ academic papers indexed and queryable in natural language
- Instant retrieval of research insights previously buried in PDFs
Context
Quiet Links is a knowledge platform for scientific research. Their team had built a growing library of academic papers, but researchers had no way to query that library intelligently — every search was manual, slow, and limited to document titles or keyword matches.
Tim Gallati, the founder, was a Python developer himself. He knew what RAG was. He estimated that building it alone would take him close to a year.
Problem
The core issue wasn't access to the documents — it was retrieval. Researchers needed to ask questions like "what does the literature say about X in the context of Y" and get an answer in seconds, not after an hour of scanning PDFs.
The secondary problem was time: Tim needed this built without waiting a year to do it himself.
Solution
A RAG backend that:
- Ingests and indexes the full academic paper corpus
- Processes natural language queries using semantic search over vector embeddings
- Returns precise, grounded answers with citations to the source documents
- Exposes clean API endpoints for the frontend to consume
Approach and stack
- My role: AI/RAG Engineer — backend architecture, vector search, LLM integration
- Collaborators: Bob Belderbos (Frontend), Tim Gallati (Product Owner)
- Stack: Python, Weaviate (vector database), OpenAI (embeddings + completion), FastAPI
- Document pipeline: PDF extraction, semantic chunking, embedding generation, index management
Process
- Audited the document corpus structure and identified chunking requirements for academic paper format
- Built the ingestion pipeline: extraction → chunking → embedding → Weaviate indexing
- Designed the retrieval layer with semantic search and context assembly for the LLM
- Integrated the API with Tim's frontend
- Validated retrieval quality against real researcher queries before delivery
- Deployed and handed off with documentation
Observable results
- 6-week delivery from zero to production-ready
- 200+ papers fully indexed and queryable
- Instant answers to natural language research questions, with source citations
- 1 year of solo development saved — Tim's own estimate
- Researchers can now query the knowledge base the way they think, not the way the filing system is organized
Learnings
Chunking strategy matters more than most teams expect. Academic papers have a specific structure (abstract, methodology, results, references) that generic fixed-size chunking destroys. Adapting the chunking to the document structure improved retrieval precision significantly.
"Without exaggeration, I learned the intricacies of developing RAG systems in 6 weeks that would have otherwise taken 6 months to 1 year on my own."
Tim Gallati — Founder, Quiet Links Library