Inspiration
Explain how BananaKart reimagines grocery delivery by focusing on sustainability, local sourcing, and AI-powered logistics instead of speed alone. Mention inefficiencies in current systems and how agentic AI can balance environmental and social impact.
What it does
Describe how BananaKart converts any recipe or user prompt into a sustainable grocery plan — using NLP to extract or generate ingredients, then sourcing locally first, routing deliveries efficiently, and minimizing carbon footprint.
How we built it
Summarize the hybrid architecture:
- Frontend (Next.js + Tailwind)
- Backend (FastAPI + Supabase)
- NLP engine (DistilBERT models for ingredient extraction + urgency classification)
- Generator (Mistral-7B via Hugging Face Inference API)
- Cache + orchestration (Supabase)
- Space + local deploy integration
Challenges we ran into
List key issues:
- Merging generator and parser behavior.
- Handling unstructured recipe text.
- Maintaining consistent JSON outputs from open models.
- Preventing Supabase import conflicts.
- Optimizing rate limits and retries for Hugging Face API.
Accomplishments that we're proud of
Include:
- A working hybrid endpoint
/analyze_or_generatethat auto-switches between parsing and generation. - End-to-end local and hosted inference via Hugging Face Spaces.
- Sustainable caching with Supabase.
- Functional front-end demo at
/cook.
What we learned
Note:
- Importance of caching and structured generation.
- Trade-offs between model interpretability and output control.
- How to unify data pipelines between Supabase and Hugging Face.
What's next for BananaKart
Add:
- Recipe generator fine-tuning on sustainability metrics.
- Monte Carlo carbon simulator.
- Route clustering and local sourcing optimization.
- Vendor partnerships and user carbon dashboards.
Built With
- javascript
- python
- supabase
Log in or sign up for Devpost to join the conversation.