At Feedcoyote, I build production ML systems that classify jailbreak and adversarial LLM content, and I architect document intelligence agents that turn invoice PDFs into validated JSON over Apache Pulsar and Apache Beam. I also engineer FastAPI microservices for high-concurrency workloads, maintaining sub-100ms P99 latency through asynchronous connection pooling.
Previously at Amazon, I developed an agentic LLM system for Amazon Athena SQL validation that reduced manual migration errors by 60%, alongside IaC-driven CI/CD that cut recovery time from three hours to 15 minutes. My work across Oracle, MAScIR, and Taager has included RAG assistants, speech recognition fine-tuning, computer vision, NLP, and LLM evaluation.
