I build intelligent agentic systems.

Building production multi-agent systems on Azure Foundry with the Microsoft Agent Framework — MCP-standardized A2A handoffs between Claude and Azure OpenAI agents.

• Orchestrated multi-agent workflows with Microsoft Agent Framework, Azure Functions for stateful coordination and Cosmos DB vector search.
• Integrated Azure SRE Agent (Preview) — KQL queries, dynamic thresholds, incident automation across AKS/App Service with human-in-loop approvals.
• RAG pipelines with Cosmos DB vector search cut triage hallucinations by 95%.
• Privacy guardrails via Azure Content Safety + Purview; PII redaction pipelines for compliant inference.
• Standardized agent integration SDKs — lifted internal team adoption 3×.

Stood up Gemini Enterprise + GPT-4/Claude 3 powered agents for real-world business workflows behind production SLAs.

• Orchestrated autonomous agents with LangChain & LangGraph — prompt chains, RAG, multi-model LLM calls.
• Modular FastAPI backend services with IAM, encryption and PII redaction aligned to enterprise governance.
• Pinecone / Qdrant vector stores + semantic retrieval to sharpen context-aware agents.
• Serverless inference (AWS Lambda, Bedrock) + CloudWatch SLO/SLI monitoring.
• CloudFormation IaC versioned model + service rollouts with rollback.

Drove AI-powered customer support and analytics — boosted efficiency and adoption across client-facing surfaces.

• NLP (classifiers, NER, NLG) + GenAI modules — full lifecycle from build to retrain.
• Multi-channel chatbots — Slack, WhatsApp, Facebook Messenger.
• Google Dataflow ETL + real-time ingestion for training/inference at scale.
• LangChain agents, RAG with LlamaIndex + vector search, AWS Guardrails.
• Fine-tuned LLMs on domain data; delivered Vertex AI NLP + CV + predictive analytics.

Computer vision, predictive models and time-series forecasting — shipped classifiers and OCR pipelines to production.

• Image enhancement + SIFT feature extraction; VGG-based classifiers with continuous retraining.
• LoRA / QLoRA fine-tuning — cut compute cost while keeping accuracy.
• S3-backed training data pipeline, DevOps CI/CD for ML services.
• Production monitoring + root-cause analysis for deployed models on AWS.

Contributed to Search ranking + YouTube recommendation systems inside Google's internal research infrastructure.

• TensorFlow neural nets on Google's internal infra; Python + SQL data pipelines.
• A/B testing + dashboards to measure model lift and inference speed.
• Cross-functional debugging of ML pipeline components.