Investment Knowledge Assistant
Deal Insights in Hours Instead of Weeks for Investment Teams
The Challenge
A private investment company's deal teams spent weeks manually sifting through internal documents, memos, and market data to prepare for investment decisions. Institutional knowledge was scattered across systems, and critical insights were often missed or rediscovered too late. The company needed a way to make their entire knowledge base instantly queryable while maintaining strict data governance.
Our Approach
We owned the full design and delivery of the RAG-based assistant. The ingestion pipeline processed diverse internal documents — memos, reports, deal data — into a structured index optimized for semantic search with metadata-aware retrieval.
The retrieval strategy combined dense vector search with re-ranking, tuned through systematic evaluation using RAGAS metrics. We iterated on chunking strategies, embedding models, and prompt templates to maximize answer relevance and citation accuracy.
The API layer was built with FastAPI, deployed on AWS with Snowflake for data warehousing and Pinecone for vector storage. Multi-model support (OpenAI, Cohere, Anthropic) allowed flexible LLM selection based on task complexity and cost.
Close collaboration with stakeholders throughout the project ensured the platform evolved with changing requirements. The result was a tool used daily by dozens of investment teams, transforming deal preparation from weeks to hours.
The Results
The competitive edge in deal sourcing often comes down to speed. By making the firm's entire institutional knowledge instantly queryable, teams could move from initial interest to informed thesis faster than before — without losing the depth that manual research previously provided.
Recuded time to insight from weeks to hours
Dozens Daily Active Teams
End-to-End Pipeline Coverage
Technologies Used
Ready to build something similar?
Let's discuss how we can apply these approaches to your specific challenges.
Get in Touch