Engineering · Full-time
Senior ML Engineer, Document AI
Own the models that read ACORD 25 certificates and multi-page ISO endorsement packets: layout detection, form number and edition date extraction, and the confidence scoring that decides when a risk engineer needs to look at a page.
- Team
- Engineering
- Location
- Delhi, India (hybrid)
- Posted
- September 15, 2026
- Compensation
- Band shared in the first call, plus stock options
- Hiring manager
- Siddharth Menon, Co-founder & CTO
The team
Where this role sits
The Document AI team is four engineers in Delhi who sit beside Risk Engineering. Every model change ships with a diff report showing which fields moved across our labeled synthetic packets, and Risk Engineering signs off on anything that changes how an endorsement edition is read. Customer files never enter training; the team generates its own packets, including the ugly ones: rotated faxes, stapled pages, and carrier forms that renumber the ISO paragraphs.
Responsibilities
What you will do
- Train and ship layout and extraction models that find form numbers, edition dates and schedule entries on scanned endorsement pages, including skewed, fax-quality scans and carrier manuscript forms.
- Own the evaluation harness: labeled ACORD 25, ACORD 28 and endorsement pages, field-level precision and recall by form, and regression gates in CI before any model reaches production.
- Design confidence calibration so uncertain fields route to the Risk Engineering review queue instead of silently passing or failing a vendor.
- Work with Risk Engineering to turn ISO form knowledge, such as how CG 20 10 04 13 differs from CG 20 10 11 85, into features, rules and test cases.
- Keep inference fast and affordable so a full packet comes back verified in about 15 seconds, even during January and July renewal peaks.
- Mentor two engineers on the Document AI team and review their experiment designs and write-ups.
Requirements
What you bring
- 5+ years building and running ML systems in production, with at least 2 years on document understanding, OCR or information extraction.
- Strong Python and PyTorch, and hands-on experience with layout-aware models or detection plus OCR pipelines, including how they fail on low-quality scans.
- Experience designing labeled datasets and annotation guidelines, and measuring label quality before trusting a metric.
- Working knowledge of serving models on AWS: GPU and CPU inference, batching, and queue-based workers.
- A habit of writing the evaluation before writing the model.
- Clear written English; your design docs are read by risk engineers as well as engineers.
Nice to have
Bonus points, not requirements
- Experience with insurance, legal or financial documents where one missed clause changes the outcome.
- Synthetic data generation for document models (we never train on customer files, so this is part of the job).
- Open-source contributions to document AI or OCR tooling.
Interviews
How we interview for Senior ML Engineer, Document AI
01
30-minute introduction with our talent partner about the role, location and compensation band.
02
60-minute conversation with Siddharth Menon about a document or extraction system you built and what broke.
03
Two-hour working session: diagnose a flawed extraction model on synthetic endorsement pages with an engineer from the team. No take-home.
04
System design on serving the parsing pipeline through a renewal peak, then a final conversation with Ananya Rao.
Apply
Send us a link and a few specifics
No cover letter needed. Tell us about something you built, shipped or fixed that relates to this role. Questions first? Email careers@pikoruasecurity.com.
Pikorua Security Private Limited is an equal opportunity employer. We consider every qualified applicant without regard to race, color, religion, sex, sexual orientation, gender identity, national origin, disability, age, veteran status or any other protected characteristic.
More roles
Also hiring
Life at Pikorua Security
Read why we started Pikorua Security before you apply.
The founding story, our six operating principles and the people you would work with are on the About page.