About the Project – Insider

Inspiration

I’m a first-year CS student at the University of Waterloo (started fall 2025) and I’ve been trading crypto since I was 16. Polymarket blew my mind when it exploded in 2024. A brand-new, crypto-native platform suddenly became the most accurate forecaster on the planet, beating polls, media, and experts during the U.S. election. It felt like the future: decentralized, global, 24/7 markets that actually reward truth. But actually trading it as a retail user sucked. Spotting an edge from news was easy; executing before the market caught up was painful. I built Insider because I wanted Polymarket’s god-tier information with Robinhood-level simplicity. One clean UI, one click, real edge.

What it does

Insider is a clean, web app that turns breaking news into Polymarket trades in seconds:

• Live news feed pulled from Google News RSS
• Google Gemini 1.5 Pro instantly maps each headline to relevant Polymarket markets and estimates a new fair probability
• Shows current market price vs. my AI-estimated fair price → the edge
• Calculates optimal Kelly bet size based on my bankroll
• One-click “Bet YES” or “Bet NO” using Coinbase Smart Wallet on Base (no seed phrases, no gas worries)
• Full Demo Mode with simulated real PnL so I can test the AI risk-free

How I built it

Solo, from scratch:

• Frontend: Next.js 14 (App Router) + Tailwind + shadcn/ui
• Web3: Coinbase Smart Wallet + Base Sepolia testnet (passkey auth with FaceID)
• Ethereum hooks: wagmi + viem
• Backend: Python FastAPI
• AI brain: Google Gemini 1.5 Pro — I feed it the headline + exact market question + current odds, and it spits out a new probability + reasoning
• News: Google News RSS + SerpAPI fallback ve

Challenges I ran into

1. Getting the LLM to reliably match messy news headlines to ambiguous Polymarket titles — ended up writing a fuzzy-matching + embedding pre-filter

2. Gemini isn’t cheap — I would’ve loved to ensemble it with Claude 3.5 + GPT-4o for better accuracy, but that would’ve been way too expensive. Had to cache aggressively and only call the model on meaningful price moves (>3%) or high-impact news

3. Coinbase Smart Wallet SDK still has rough edges on some devices

4. Kelly Criterion can recommend insane bet sizes when the model is very confident — added a “fractional Kelly” slider so I don’t blow up in real life

Accomplishments I’m proud of

• Built and shipped a full-stack production-grade dApp completely by myself as a first-year university student
• First time using account abstraction (ERC-4337) and it genuinely feels like Web2 onboarding
• The UI is something I legitimately use every day — no crypto jargon, just clean and fast

What I learned

• Modern Web3 tooling has finally reached the point where crypto UX can beat traditional fintech
• LLMs are insanely good at probabilistic reasoning if you give them clean context
• Base is hands-down the best L2 right now for consumer apps — sub-cent fees and 2-second finality

What’s next for Insider

• Telegram/Discord alerts for high-edge signals
• Mainnet deployment with real money
• Ensemble of multiple models (Gemini + Claude + Grok) to reduce hallucinations
• Portfolio dashboard with performance tracking
• Eventually open-source parts of it so others can build their own strategies

Built With

• base-sepolia-testnet
• coinbase-smart-wallet
• fastapi
• google-gemini-api
• googlenews
• next.js
• python
• react
• tailwind-css
• tanstack-query
• typescript
• viem
• wagmi