I've built production features for Sprih's hierarchical multi-agent platform, enabling 20+ end-to-end workflows for research, drafting, review, and report generation. I developed sandboxed sub-agents, S3-backed workspaces, terminal execution tools, and human-in-the-loop Word document review systems.
I reduced effective prompt context by about 65% through summarization middleware and built evaluation harnesses for ESG research agents across Azure Foundry and GCP Vertex AI. My work includes LangGraph, DeepAgents, LangSmith, OOXML parsing, content hashing, and production-focused testing.
At Point9 AI and Vidysea, I built FastAPI and MongoDB services, Celery task orchestration, and LangGraph-powered data pipelines processing 50K+ records. I also created resilient LLM orchestration with fault recovery, review gates, schema validation, and quarantine tracking.
Previously at CountryBean, I integrated RAG chatbots, persistent memory, fine-tuned LLaMA with QLoRA, and built tool-using agents. I also build end-to-end ML systems, including churn prediction pipelines and a LangGraph-based product recommendation engine.
