Hackathon Portal
AI Tinkerers - Paris
Sign in

Paris Vibe-a-thon: “Build a One-Person Unicorn” — November 15, 2025

Sign in
Engineering Lead at Stealth AI Startup - Stealth AI Startup
Own AI-native architecture and backend pipelines remotely with no equity gimmicks.
All Jobs
Final round winners have been announced. View Results
Team

Diggyy Fresh

Project Concept

Building FreshFlow AI - an intelligent inventory agent for grocery stores that uses reinforcement learning and LLMs to reduce food waste by 40-60% while improving fresh product availability.

Entry

Status: Submitted

Last saved: November 15 at 5:53 PM CET

View Entry

Team Roster

Message board not available for this team yet.

KP Akpomiemie Team Lead

AI engineer at Emlyon
Solo Developer - KP: I was responsible for all aspects of the Diggyy Fresh project from conception to deployment. I designed the system architecture, defined the reinforcement learning problem formulation (state/action spaces, reward function), and integrated the LLM component for explainability. I used **Claude Sonnet 4.5 via Claude Code** as my primary development tool, which generated approximately 80% of the initial codebase including the Gymnasium environment implementation, PPO training pipeline using Stable-Baselines3, and the Streamlit dashboard with Plotly visualizations. I leveraged Claude Code's autonomous coding capabilities to rapidly prototype the project structure, debug integration issues, and optimize the reward function - tasks that would typically require a full development team. On the AI infrastructure side, I implemented **Ollama's API (localhost:11434)** to run **Llama 3.2 3B** locally for demand signal interpretation and decision explanations, avoiding cloud API costs while maintaining fast inference. I built the synthetic data generator when the Instacart dataset became unavailable, created the Docker deployment configuration, and designed the entire demo narrative flow. Claude Code specifically helped me troubleshoot the PyTorch version conflicts, create modular Python code with proper error handling, and generate comprehensive documentation - essentially acting as an AI pair programmer that compressed weeks of solo development into a 4-hour hackathon sprint. All version control, testing, and deployment decisions were mine, with Claude Code accelerating execution velocity.
I’m a masters student building my skills in AI engineering.
I’m looking to learn and collaborate with people working on reinforcement learning, AI agents, Applied LLMs, and RAG.
I’m currently working on RL projects.