-
-
Landing page
-
Lsnding page 1.1
-
Landing page 1.2
-
Register page
-
Login page
-
Dashboard page
-
Dashboard page 1.1
-
AI Chat page
-
About page
-
About page 1.1
Inspiration
I used to watch my relatives make investment decisions based on WhatsApp forwards and random YouTube videos. Not because they were careless—but because every "smart" financial tool either buried them in charts they couldn't interpret or gave AI-generated advice that wasn't based on what the market was actually doing right now.
That gap frustrated me. Why should grounded, real-time financial intelligence be locked behind expensive professional platforms?
I built MarketPulse because I wanted a tool where someone could simply ask, "Is Bitcoin risky today?" and get an honest, data-backed answer. As a second-year CS student, I wanted to prove that reliable AI doesn't have to be complicated—it has to be grounded.
What it does
MarketPulse is an AI-powered financial intelligence platform that analyzes stocks and cryptocurrencies using real market data stored in MongoDB.
Users can ask questions like "Is Bitcoin risky right now?" and receive structured responses including risk level (HIGH/MEDIUM/LOW), trend direction, confidence score, and a short expert-style insight—all generated using current market data retrieved from the platform's database.
How I built it
The frontend was built with React and Vite, styled with a Bloomberg Terminal-inspired cream and gold aesthetic. The backend uses FastAPI and MongoDB Atlas with JWT authentication.
The AI layer uses a LangGraph ReAct agent powered by Llama 3.3-70B through Groq. Two custom tools—get_crypto_prices and get_stock_prices—query MongoDB before generating market analysis.
An APScheduler background job refreshes 4 cryptocurrencies and 4 stocks from CoinGecko and Yahoo Finance every 15 minutes, eliminating the need for live API calls during chat.
Challenges I ran into
API Rate Limits: Alpha Vantage's free tier (25 requests/day) became unusable in production. I switched to Yahoo Finance, which provides free market data without API keys or strict limits.
LLM Grounding: Early versions sometimes answered from training knowledge instead of current market data. I solved this with a strict system prompt and tool-based workflow that enforces database retrieval for market questions.
MongoDB SSL on Python 3.14: Render's default Python 3.14 environment introduced SSL compatibility issues with PyMongo. The issue was resolved by pinning the runtime to Python 3.11.
Cold Start UX: Render's free tier introduces startup delays after inactivity. I mitigated this with frontend caching to make repeat visits feel instant.
Accomplishments that I'm proud of
Building and deploying a complete AI application—with authentication, automated data pipelines, a grounded AI agent, and a polished user experience—entirely as a solo developer in under two weeks as a 2nd year undergraduate.
I'm especially proud that the project is fully deployed and solves a real-world problem rather than remaining a prototype.
What I learned
The hardest part of AI applications isn't the model—it's the data pipeline.
Getting fresh, reliable information into the agent's context at the right moment, without excessive latency, rate limits, or inaccurate responses, is the real engineering challenge.
This project also taught me that deployment exposes issues that local development never reveals. Shipping to production is where the most valuable learning happens.
What's next for MarketPulse
Future plans include portfolio risk analysis, personalized watchlists with alerts, financial news sentiment analysis, and multi-agent LangGraph investment research workflows.
My goal is to make trustworthy, data-driven financial intelligence more accessible to everyday investors.
Log in or sign up for Devpost to join the conversation.