FinFAST.ai: faster retrieval and cheaper AI for an economic research platform
A research platform where analysts query large economic datasets and read AI-written daily reports. I made it fast, made its AI answers more accurate, and cut what the AI cost to run.
- Context
- Built at Oxbridge Economics as main developer
- Role
- Main developer, full stack
- When
- 2024 to 2026
Results
| Measure | Before | After | Change |
|---|---|---|---|
| Typical data retrieval | ~10 s | ~1.5 s | 6.7× faster |
| Some queries | ~2 s | 0.5 s | 4× faster |
| Daily LLM token cost | 100% | 20% | −80% |
| RAG retrieval accuracy | baseline | +70% | +70% |
The problem
Queries against large datasets took around 10 seconds, and some sources timed out entirely. The daily multi-agent report system was expensive to run, and the retrieval behind its answers often surfaced the wrong material.
What I did
Retrieval performance
Consolidated chatty requests, indexed the MongoDB keys the queries actually filter on, optimized search, added caching, and paginated the APIs. Sources that used to time out now return in about 1.5 seconds.
Cheaper daily reports
Owned the daily multi-agent economic-report system and restructured its data-collection prompts so the model provider's cache is reused, reducing daily token cost by 80%.
Better answers
Redesigned the retrieval-augmented generation (RAG) algorithm behind research and discovery, improving retrieval accuracy by 70%.
Product and platform
Built the FinAnalytics dataset-aware chat workspace, refined the DataHub dataset catalog, added Google sign-in, and deployed frontend and backend across AWS and Azure with CI/CD.
How it fits together
1
Data sources
Economic datasets, news, market feeds
2
Pipelines
Topic clouds, sentiment, scheduled collection
3
Retrieval
Indexed MongoDB, cache, paginated APIs: ~1.5 s
4
AI layer
RAG +70% accuracy, multi-agent reports −80% cost
5
Web app
Next.js workspace, Google OAuth, AWS + Azure
Stack
- Next.js
- React
- TypeScript
- Python
- MongoDB
- AWS Amplify, CloudFront, API Gateway, Lambda, EC2
- DynamoDB / RDS
- Azure AI Foundry, Cosmos DB