Hello, I’m Anusha.
ANUSHA ACHARYA
My work spans the AI development cycle: training models, building data collection platforms, and using evaluations and traces to improve how systems perform.
Studying how agents negotiate.
I build environments to study multi-agent negotiation. We collect interaction data and decision traces, compare models, and investigate what makes an evaluation meaningful when agents interact with each other.
Founding engineer.
As the first engineer at Pipelines, I built a platform supporting the full AI R&D cycle: multimodal data collection, generation, workflow orchestration, and model evaluation. That included synthetic data pipelines and three layers of evaluation: programmatic checks, LLM judges, and human ratings.
Midstream Health, San Francisco.
At Midstream, I lead document extraction and contract ingestion, and work on post-training models for extraction. I use execution traces and human feedback to build evaluation datasets, understand failures, and improve AI baselines through repeated experiments. The goal is reliable structured data that financial agents and workflows can use.
Meta and Bloomberg.
At Meta, I worked on model training, improving Instagram’s follow prediction through better training data and a weighted quality score. I also built voice and NLP integrations, agentic workflow automation, and product infrastructure. Before that, I worked on logging and data quality at Bloomberg.
UMass Amherst.
A master's in computer science, and a term grading neural networks for COMPSCI 682, which is the fastest way there is to find out how well you actually understand a thing. Before it, a first degree in computer science and engineering from Visvesvaraya Technological University, finished in 2019, and two years at Sandvik in Bengaluru building reporting and a model that predicted when mining loaders would break.
English · Hindi · Kannada
अनुशा