Machine Learning Engineer

San Francisco, United States

Full-time

About the role

Reducto trains vision models that read complex documents the way a human would, and turns them into inputs an LLM can actually use. You would join an eight-person machine learning team, reporting directly to co-founder and CTO Raunak Chowdhuri, and own models end to end from research through production. The team runs a mix of smaller in-house models and larger open-source ones, and ships against real enterprise workloads: FAANG customers, top trading firms, and the AI teams building on top of Reducto.

You will feel at home here if you have shipped a model yourself, all the way to users, and never handed it off half-finished.

What You'll Do

Train and deploy. Build new state of the art models for parsing and interpreting unstructured data.
Experiment. Try novel techniques to improve LLM accuracy, and build the tooling to know whether they worked.
Own the pipeline. Build data pipelines, evaluate model performance, and integrate models into the product.
Work with founders and customers. Shape product direction and engineering strategy directly, not through a layer.

About Reducto

Reducto is a Series B team in San Francisco, founded in 2023 by Adit Abraham (CEO) and Raunak Chowdhuri (CTO), who met at MIT. Chowdhuri had published computer vision papers with over 100 citations before finishing high school. The company went through YC Winter 2024 and has raised $108M from a16z, Benchmark, First Round, BoxGroup, and YC.

The machine learning team is eight people and reports to the CTO. It runs a mix of in-house models and larger open-source ones, and an ML tech lead role is expected to be filled around the new year. Engineering is in the office five days a week near Union Square and Market, which is intentional at this stage.

Culture signals that hold across every role here: high agency and self-direction, no settling for 90%, and attention to detail that matters concretely. In document parsing a comma versus a period turns $20,000 into 20.000, so sloppy work is a real red flag. Engineers are expected to hold technical conversations with enterprise customers, not just stay in the code.