Ask Your Own Docs
Internal knowledge assistant using RAG to surface answers from 5 years of project documentation, with citation and confidence scoring.
Context
Meridian had 5 years of project documentation scattered across Notion, Google Drive, and old wikis. New consultants spent days searching for precedents. Senior staff answered the same questions repeatedly. They wanted to "ask their own docs" but needed it done carefully.
Constraints
- !Sensitive client information in documents
- !Must cite sources—no hallucinations acceptable
- !Team skeptical of AI accuracy
- !Limited AI/ML expertise in-house
Approach
We implemented a RAG (Retrieval-Augmented Generation) system with strict guardrails. Documents were chunked and embedded with metadata. The interface showed confidence scores and always linked to source documents. We built evaluation datasets to measure accuracy before launch.
What We Built
- Document ingestion pipeline (Notion, Drive, Wiki)
- Vector database with metadata filtering
- Chat interface with source citations
- Confidence scoring and fallback to search
- Evaluation framework with test cases
Outcome
The system achieved 89% accuracy on the evaluation set before launch. Adoption was faster than expected—the citation feature built trust. Time-to-answer for common questions dropped from hours to minutes.
Tech Stack
“The citations were key. People trust it because they can verify. Our new consultants are productive weeks earlier now. The evaluation framework means we can measure if it's actually working.”
What We'd Do Next
- Client-specific knowledge bases
- Integration with project management
- Automated document summarization
Related Projects
Tables, Not Clicks
Bistro Laurent
Restaurant site rebuild focused on mobile reservations, resulting in 3x online bookings and reduced no-show rate through smart reminders.
From Spreadsheets to System
FlowState (Internal)
Internal operations dashboard consolidating 5 spreadsheets into real-time views, saving 10+ hours weekly and eliminating data sync errors.
Have a similar challenge?
Let's discuss how we can help you achieve similar results.