SlipstreamJobs tracks this role from the company's public career site. Apply directly on the employer's site.
Katalyze AI is building AI systems to extract and structure data from pharmaceutical manufacturing records. You will own the machine learning systems that read scanned batch records, certificates of analysis, deviation records, and related documents—turning unstructured pages into structured, source-cited data that scientists and engineers can trust.
In this role, you will:
- Train and evaluate models that read scanned and digital manufacturing records, working with OCR and language models to extract accurate information
- Build data pipelines that transform each model read into structured data with full source attribution
- Measure accuracy on real customer documents and systematically close gaps you identify
- Design agent workflows that retrieve, reason over, and cite records across multiple systems
- Build evaluation sets in collaboration with domain experts and run them on every model change
- Optimize for accuracy, latency, and cost while mentoring other engineers on the AI team
- Work closely with deployment and customer-facing teams to ensure models perform reliably in production
- Own systems from prototype through production deployment
You'll be part of a small, focused AI team working on a high-impact problem in a regulated industry. The role is based in the New York City office with remote flexibility.
REQUIREMENTS:
- Five or more years building machine learning systems in production
- Experience with document understanding, OCR, or language models
- Strong Python skills and a disciplined approach to measurement and experimentation
- Experience building retrieval systems and evaluation frameworks for LLM-based products
- Proven track record taking models from prototype to production and maintaining reliability
- Clear communication about technical trade-offs and demonstrated care for end users
NICE TO HAVE:
- Experience in life sciences or another regulated industry
- Published work in document AI or information extraction
- Experience deploying models in on-premises or validated environments
