Work — Sanjana Londhe

Case studies

Agent Simulation

testing voice and chat agents before real users do

60–70% reduction in testing effort through automated scenario generation.
50% cut in pre-production validation time.
2× faster issue detection before launch.

FutureAGI · 3 teams running simulations in beta without engineering supportView case study

Workbench

version control for prompt engineering

2× faster prompt iterations, with one-click version restore.
500-row test runs with no code, covering full input range.
One-click eval let PMs decide on scores, not intuition.

FutureAGI · Adopted by ML engineeringView case study

Evaluation

the quality layer for AI outputs

70+ built-in eval templates across quality, safety, RAG, agents, audio, and image.
One eval template definition runs identically across datasets, live tracing, simulations, and CI/CD.
Feedback and retune turned evaluator disagreement into a training signal instead of a dead end.

FutureAGI · Shipped across FutureAGI’s platformView case study

Shipped projects

Orin

Captures and plans the in-between tasks that never made it onto the big roadmap. Designed, built, and shipped solo, and used daily to run my own task and project workflow.

LiveLive project