At Metasys Consulting, I built a legal document search system using LangChain, LlamaIndex, and Pinecone to search more than 500K documents with p95 retrieval latency below two seconds.
I trained a BERT document classifier with PyTorch and LoRA, deployed it on AWS SageMaker, and built AI agents to extract information and process data. I also reduced OpenAI API costs by 42% through caching, shorter prompts, and task-specific model selection.
At Nextdev Labs, I added GPT-4o features to analytics dashboards and built a chat interface for asking questions about MongoDB data. I also worked on model-behavior alerts and deployed services on AWS EKS using Kafka and Redis.
In my projects, I built Sagefold, a private document Q&A platform with source citations, and a multi-agent playground connected through a WebSocket hub. I also created BloomOps, a machine-learning workflow for training, tracking, and serving an Iris classification model.
