1st Place · Atrato Financial Hackathon, 1st Edition
FinanceAI
AI assistant that answers questions about financial products and routes users to the right expert, with embedding-based cost optimization.
- My role
- Team member
- Team
- 4 people
- Event
- Atrato Financial Hackathon, 1st Edition · Atrato
- Year
- 2024
The problem
Atrato wanted an AI virtual assistant that automates workflows by answering customer questions about its financial products.
What I built
- A question-answering assistant that also matches users to the right problem-solver from an uploaded employee database.
- Real-time API cost metrics, with costs reduced through embeddings and vectorized datasets.
- SQL-injection prevention on prompt input and hate-speech filtering on LLM output.
- A TruLens feedback loop to evaluate and improve answers.
- A responsive Streamlit app with light/dark mode, deployed on Streamlit Cloud.
Results
- 1st place at Atrato's Financial Hackathon, 1st Edition.