ML engineering project
FinFluence
FinFluence is an AI-powered financial literacy platform that turns short-form video education into a tokenized economy where both creating and consuming knowledge gets rewarded with real cryptocurrency. **The Problem:** Gen-Z spends over 3 hours daily scrolling TikTok, Instagram Reels, and YouTube Shorts, yet less than 20% feel confident making basic financial decisions. Financial advice online is fragmented, inconsistent, and designed for clicks rather than education. **Our Solution:** • **AI-Validated Content** - Custom sentence transformer models verify that every reel is genuinely about finance. No spam, no off-topic content—just quality educational material • **Tokenized Rewards (FINF)** - Custom Solana SPL token. Watch educational content? Get paid. Create quality videos? Get paid more. Real crypto in your wallet based on genuine engagement • **Smart Feed** - Familiar short-form swipeable interface optimized for financial education with real engagement tracking and heuristic bot detection • **Non-Custodial Security** - Users always control their own assets through Solana Wallet integration **Architecture:** • **Frontend** - React + TypeScript + Vite, Tailwind CSS, Solana Wallet Adapter, Three.js for 3D token visualization, Framer Motion animations • **Backend** - Python FastAPI with async processing, modular routers, Pydantic validation, JWT auth, Docker deployment • **AI Pipeline** - Sentence transformers compute semantic similarity between uploaded content and financial topics. Real-time validation during upload, multi-dimensional content scoring • **Blockchain** - Custom Solana SPL token with smart treasury system tracking engagement, calculating fair reward splits, and executing on-chain transfers • **Data Layer** - MongoDB with GridFS for efficient video storage (handling 16MB+ files), indexed queries, user/reel/engagement schemas **Key Challenges Solved:** • **AI Latency** - Async processing queue: uploads accepted instantly, validated in background, creators notified within 30 seconds • **Tokenomics Balance** - Dynamic reward system considering retention rate, engagement depth, creator reputation, and network effects to prevent bot farms • **Video Streaming** - GridFS chunking with proper byte-range request handling for smooth playback **Results:** Production-ready multi-service platform deployed in 36 hours. AI content validation, blockchain integration, video streaming—all orchestrated with Docker Compose. **Tech Stack:** Built with **FastAPI**, **React/TypeScript**, **Sentence Transformers**, **Solana SPL**, **MongoDB GridFS**, **Docker**, **Three.js**, and **Tailwind CSS**.
Problem
Short-form financial content is easy to consume but difficult to verify, reward, and organize around genuine learning.
My role
Built the multi-service hackathon prototype with the team, connecting the FastAPI pipeline, React experience, content validation, storage, and token workflow.
Approach
Validate uploads with sentence-transformer similarity, stream video from MongoDB GridFS, and reward qualified engagement with a Solana SPL token.
Outcome
Delivered an integrated multi-service prototype during the 36-hour SwampHacks XI build and documented the latency, streaming, and token-design lessons.
Implementation
- Ran semantic content checks asynchronously after upload.
- Stored video with MongoDB GridFS and served a short-form React feed.
- Connected engagement calculations to a non-custodial Solana token flow.
Evaluation
- Validated the end-to-end upload, classification, streaming, and wallet flow during the hackathon.
- Used content-relevance scoring and heuristic engagement checks rather than claiming educational efficacy.
Results
- Produced a working multi-service prototype in a 36-hour hackathon window.
- No user-adoption or learning-outcome metric is claimed.
Constraints
- Video validation could not block the upload experience.
- Large media required chunked storage and byte-range delivery.
- Rewards needed protections against shallow engagement and automated abuse.
Tradeoffs
- Token rewards can motivate participation but also create incentives for gaming.
- Background validation improves responsiveness while delaying final publication decisions.
Next improvements
- Run educator-reviewed quality evaluations and longitudinal learning studies.
- Strengthen fraud detection and make reward policy easier to audit.