AI/ML Engineer building production-grade machine learning systems and agentic AI pipelines that drive measurable business outcomes.

Currently an AI/ML Engineer at BNY (New York), working on financial ML models, anomaly detection, and LLM-powered retrieval systems. I hold NVIDIA Generative AI Professional and NVIDIA Agentic AI Professional certifications.

• 🎯 Current Focus: Agentic AI systems, multi-agent orchestration, production RAG pipelines, and LLMOps
• 🔭 Working On:
  • Graph-based RAG knowledge assistants with multi-hop reasoning
  • Multimodal agentic security investigation platforms
  • Multi-agent AI travel planning systems with hybrid retrieval
• 🎓 M.S. Computer Science (AI/ML) — SUNY Binghamton (May 2025)
• 💡 Interests: Autonomous agents, LLM fine-tuning, structured outputs, production ML infrastructure, prompt engineering at scale

• Trained XGBoost classification models achieving 91% accuracy, supporting 35% reduction in manual review effort
• Built PyTorch neural network for transaction anomaly detection, improving accuracy 18% over baseline
• Developed feature engineering pipelines reducing manual data prep time by 40%
• Deployed agentic AI pipeline using LangChain + RAG for LLM-powered document retrieval, reducing analytics research time ~30%
• Containerized ML models with Docker; supported Azure ML deployments with MLflow monitoring, reducing release cycles from biweekly to weekly

• Built collaborative filtering recommendation engine (Python, TensorFlow) on 1M+ transactions, driving 24% conversion improvement
• Deployed real-time XGBoost pricing model processing 50K+ products hourly, delivering 16% annual revenue growth
• Executed EDA and feature engineering on 2M+ customer records using Pandas/NumPy, improving model metrics 19%
• Deployed SageMaker models with Docker containerization; built Tableau KPI dashboards for stakeholder reporting
• Integrated ML models into AWS microservices via REST APIs for real-time inference with performance monitoring

Tools: LangGraph, Neo4j, OpenAI, FastAPI, Langfuse

Entity-extraction and graph-ingestion pipelines over financial filings with LLM-guided traversal, improving multi-hop recall 38% vs. dense-only retrieval. Instrumented full pipeline observability with Langfuse tracing and RAGAS evaluation harness comparing fixed, semantic, and late-chunking strategies.

Tools: FastAPI, Next.js, Mistral, MCP, OCR, SSE

Multimodal agentic pipeline automating OCR-based evidence extraction, attack-graph generation, and root-cause analysis. Built structured MCP tool registry with LLM function-calling and strict JSON outputs for type-safe agent execution.

Tools: LangGraph, AWS Bedrock, GPT-4o-mini, pgvector, Neo4j

Multi-agent AI travel planner orchestrating planner, researcher, and packager agents for multi-step itinerary generation. Optimized hybrid RAG pipeline with chunking strategies and vector store indexing via Supabase pgvector, boosting route coherence 40%.

Last Updated: June 2026 | Credits: Ankit Patil