Quandary AI

Perplexity for Finance. Quandary AI with real-time market intelligence and AI research capabilities.

Comment

Inspiration

I’ve been genuinely impressed by what Perplexity can do—it’s quickly become my go-to tool for daily research. Staying on top of financial news and managing portfolio risk can be overwhelming and expensive, often requiring costly research services. The goal behind this project is to bring that kind of personalized insight and exploratory analysis directly to users, so they can assess their portfolios, experiment with different strategies, and better understand their growth potential—all in one place.

What it does

Quandary AI allows users to upload their portfolio data in Excel format. The system parses and processes this data to generate key performance indicators (KPIs), which are then analyzed using numerical computation libraries such as NumPy for in-depth risk assessment. Leveraging the capabilities of the Sonar Pro large language model (LLM), Quandary AI delivers comprehensive portfolio insights including:

  • Portfolio Health Scoring
  • Risk Assessment and Diversification Analysis
  • AI-Powered Investment Recommendations
  • Scenario Simulation and What-If Analysis
  • Real-Time Market Data Integration
  • This combination of quantitative analysis and AI-driven research enables users to make informed, data-backed investment decisions with ease.

How we built it

  • Frontend - React + Vite
  • Backend - Python + Flask
  • Firebase for authentication
  • Vercel for frontend deployment ( CI/CD is done for every push code will be deployed)
  • Google Cloud Run for Backend deployment

Challenges we ran into

  • Vercel was my option to deploy both frontend and backend, but backend was not dpeloying and i don't see any new logs, so i have to go for the google cloud.
  • With Vercel the proxy to the google cloud was not working, bit panicked with that as it was working fine in the local. But able to figure the proxy configuration when deploying into the produciton.
  • Ran into the parsing issue multiple times as the data generated by the api is not deterministic. Managed to find the pattern and work along with that. ## Accomplishments that we're proud of
  • Able to write the targetted prompt and generate the KPIs out of it. ## What we learned
  • Vercel deployment - First time
  • Video recording a demo )personally) ## What's next for Quandary AI
  • Need lot of improvements in the UI with Market reasearch
  • Fine tuning the prompt
  • History being stored information
  • Scenario Simulation / What-if Analysis
Share this project:

Updates