Show HN: Navital Gym – open environments for testing AI agents on Indian tasks

Category: ai-ml

Tags: ai-agents, benchmark, evaluation, india, multilingual, deterministic-grading

Score: 7.0/10 (Innovation: 7, Technical: 7, Documentation: 8, Utility: 6)

Navital Gym provides deterministic, rule-based benchmark environments for evaluating AI agents on India-specific tasks such as trip planning, GST invoicing, UPI payments, form filling, and multilingual customer support. Its key differentiator is exact, recomputed grading without LLM judges, honest handling of impossible requests, and hidden fields that never leak to the agent, all implemented in pure Python with adapters for popular agent eval frameworks.

Target audience: AI/ML researchers and engineers evaluating agents on domain-specific and multilingual tasks, particularly those targeting Indian markets

Repository: https://github.com/Navitallabs/navital-gym · Python · Apache-2.0 · 1 stars

View on Hacker News