Senior AI Engineer SIRO
Architected centralized asynchronous LLM infrastructure using Python asyncio and Redis-backed distributed queues to support concurrent multi-agent workloads. Implemented dynamic token-bucket throttling and request orchestration to improve API throughput, mitigate HTTP 429 rate-limit errors, and handle traffic spikes.
Designed a three-stage agentic RAG pipeline with dynamic semantic chunking for processing 100+ page technical documents against a corpus of more than 6,000 historical documents across five therapeutic areas. Reduced complex regulatory reviews from days to minutes through a double-pass LLM validation layer with deterministic scoring and auditable outputs.
Built concurrent, context-aware agentic pipelines to validate clinical protocols against more than 4,500 documents and 50,000 pages of heterogeneous content. Reduced global adaptation reviews from three weeks to under five minutes and lowered localized regulatory consulting costs.
Engineered a production multi-agent library of nine specialized agents to ingest unstructured clinical PDFs, parse relational tables and flowchart decision trees, and convert source content into deterministic conditions within a queryable knowledge base.
