01 /
Context
Machine learning engineering across document intelligence, semantic search and LLM evaluation, taking models from experiment to production pipelines on AWS for legal and business use cases.
02 /
What I owned
- Built a SageMaker-based document AI system with LayoutLM + BERT, improving F1 by 18% on 10K+ legal documents.
- Deployed an OCR + LLM pipeline on Amazon EKS, cutting review time to under 10 minutes per document.
- Built a FAISS-based semantic search system with 300 ms retrieval, reducing analyst lookup time by 60%.
- Benchmarked GPT-4 against Claude for contract analysis, reducing inference cost by 35%.
- Productionized AWS ML pipelines for batch and real-time workflows.
03 /
Outcomes
- F1 improvement
- 18%
- Less analyst lookup time
- 60%
- Lower LLM inference cost
- 35%
- Review time per document
- < 10 min