Show HN: OpenDecision – a 400M zero-shot model makes local decisions, plays Doom

Category: ai-ml

Tags: zero-shot-classification, local-inference, python, fastapi, nlp, multimodal

Score: 6.8/10 (Innovation: 6, Technical: 7, Documentation: 8, Utility: 6)

OpenDecision wraps a zero-shot classification model behind a typed decision API for text, documents, and images, exposing Choice, Noul, Score, and Relation question types. It runs locally on CPU, ships a FastAPI server, and includes document retrieval and image demos like an alarm-triggering bear detector. The project is a well-documented, practical abstraction over existing models rather than a new modeling paradigm.

Target audience: backend devs, data engineers, ML engineers

Repository: https://deepanwadhwa.github.io/OpenDecision/ · Python · Apache-2.0 · 106 stars

View on Hacker News