I build & explore limits of enterprise AI systems. I also like to write about them.
Applications that use frontier models to solve hard problems; full DevOps lifecycle.
12 years from consulting (PwC M&A) to AI deployments.
Enterprises keep misreading what a model actually is: top-down decisions wrap it in bespoke orchestration or layer it onto existing software as a mere glorified autocomplete. A case for the opposite — a frontier model with filesystem access, in a structure that governs by design.
Read the paper →Multi-agent systems with ReAct methodology. RAG pipelines with HyDE retrieval. Frontier LLM integrations with zero data retention compliance.
Python backends, TypeScript frontends, REST APIs. End-to-end implementation from prototype to production deployment.
Full infrastructure ownership across AWS, Azure, GCP. Containerized deployments, CI/CD pipelines, isolated environments. HIPAA-compliant when required.
RAG pipeline for supplier contract review. Multi-document synthesis, clause extraction, semantic search.
90% reduction in review timeQueries WHO/EMA/clinicaltrials.gov APIs, cross-references proprietary regulatory database. Outputs submission-ready documents.
4+ hours → 8 minutes per cycleFAISS + S3 RAG pipeline with HyDE retrieval. Legal contract review and multi-document comparison workflows.
Enterprise-wide deploymentUnified MySQL, Snowflake, and Tableau into a conversational layer. Employees query business data through natural language.
Deep system integration, one interface