Research
Understanding AI progress.
Evals for AI-native Services Companies
When experts disagree there is no answer key, so let them review the work and turn their judgment into evals.
Don’t vibe code your software clones
Why high-fidelity software replicas demand more than surface-level generation.
Evaluating Sovereign AI
Testing Sarvam models on multilingual ecommerce support tasks with tools, policy constraints, and backend state.
Tech Support Environment
The Tham Luang Cave.
SalesforceBench
Can agents actually work inside a simulated Salesforce org?
Editing is Hard
Can LLMs edit PPTX reliably?