Inspiration Every day, millions of retail investors drown in an ocean of financial news — hundreds of articles, conflicting opinions, and market updates — with no way to know what actually matters to them. We were frustrated watching friends miss key market moves simply because they couldn't filter the noise. Bloomberg and Moneycontrol serve everyone the same feed, regardless of whether you hold Infosys or are tracking clean energy stocks. We asked: what if your financial news feed knew you as well as Netflix does? That question became FinWin.
What it does FinWin is a personalized financial intelligence feed powered by AI. Instead of showing every investor the same headlines, FinWin:
Scores and ranks news articles by relevance to your personal portfolio, watchlist, and sectors of interest Generates one-tap AI summaries so you understand any article in under 10 seconds Delivers smart alerts when breaking news directly impacts stocks you hold Shows a sentiment dashboard — visual heatmaps of market mood across sectors Learns continuously from your reading behavior to sharpen recommendations over time How we built it We built FinWin as a full-stack web app deployed at finwin-feed.vercel.app:
Frontend: React.js + TypeScript + TailwindCSS, deployed on Vercel Backend: Node.js + Express with REST APIs and WebSocket support for real-time updates AI Layer: OpenAI GPT for article summarization, custom NLP pipeline for relevance scoring and sentiment analysis, LangChain for agentic workflows Data: Financial news aggregated via RSS feeds and third-party news APIs, enriched with market data Auth & Storage: Supabase for user profiles, preferences, and behavior tracking Challenges we ran into Relevance scoring at scale: Building an NLP model that meaningfully ranks articles against a user's portfolio — without hallucinating connections — required careful prompt engineering and multiple iterations. Real-time pipeline: Pulling, processing, and scoring hundreds of articles per hour while keeping the feed snappy was a significant infrastructure challenge. Cold start problem: New users have no history, so we designed an onboarding quiz to capture risk profile, sectors, and interests upfront to bootstrap personalization from day one. Noise vs. signal balance: Tuning the algorithm to surface genuinely important news without over-filtering or burying relevant stories took extensive testing. Accomplishments that we're proud of Shipped a fully functional, deployed MVP — not just a prototype or mockup Built an end-to-end AI relevance pipeline that personalizes a feed in real time Achieved an estimated 80% reduction in irrelevant content in user feeds compared to generic portals Designed an intuitive UI that makes complex AI-driven insights feel effortless and accessible to everyday investors Completed the entire product — ideation to deployment — within the hackathon window What we learned Personalization is a product problem as much as a technical one — understanding what the user cares about is harder than building the engine that serves it LLM-based summarization works remarkably well for financial content when given tight, domain-specific prompts Real-time data pipelines require far more error handling and fallback logic than anticipated The biggest UX win was simplicity — investors don't want dashboards, they want a clean, scrollable feed that just works What's next for Finwin Voice briefings — a 60-second AI-narrated morning market brief tailored to your portfolio Broker integrations — connect Zerodha, Groww, or Upstox to auto-sync your actual holdings Predictive signals — move from reactive news to proactive alerts based on earnings patterns and historical sentiment cycles Community layer — see what stocks peers with similar profiles are reading about, without the noise of social media Multi-language support — Hindi, Tamil, and other Indian languages to reach the next 200 million retail investors B2B API — offer the personalization engine as a white-label product to fintech platforms and brokerages
Log in or sign up for Devpost to join the conversation.