Project 1
* Built real-time GTM pipelines to process 250K+ visitor events/day across web and campaign channels, capturing clicks, scroll depth, dwell time, referral source, and repeat sessions, then joining those signals with CRM/account data for a unified engagement view.
* Designed enrichment and lead-scoring workflows using behavioral and CRM signals to prioritize high-intent accounts, helping move qualified-lead-to-opportunity conversion into the 15%–25% benchmark range.
* Automated follow-up routing, segmentation, and campaign triggers based on visitor intent, session history, and product/business rules, routing qualified inbound leads and supporting request-to-book conversion.
Project 2
* Orchestrated and deployed scalable AI agents using LangChain, LangGraph, and RAG across Azure and AWS, resulting in a 25% boost in sales conversions through intelligent automation and customer interaction flows
* Led the development of MVPs and POCs for AI-powered sales agents, including voice, text, and email-based interfaces, achieving a 35% increase in customer engagement
* Developed a real-time voice assistant (Phone Agent) using Twilio, WebSocket, Flask, AWS Lambda, S3, DynamoDB, and Claude 3.5 Haiku via Bedrock, integrated with LangChain-backed RAG pipelines, achieving a 10% reduction in response latency
Project 3
* Developed multi-agentic orchestration application with 7+ agents in production using Google ADK framework with A2A communication and LangChain, connecting with RAG, multiple MCP tools, and vLLM for inference
* Trained a Text-2-SQL model using Supervised PEFT (LoRA) on 4 H-100 GPUs using PyTorch Distributed Data Parallel Technique (DDP), fine-tuned bge-large-en-v1.5 embedding model & evaluated using LLM-as-a-Judge Framework
* Performed Domain Adaptive Pre-training (DAPT) & prepared pre-training and fine-tuning data to train foundational LLM on energy domain using Llama-3.1-70B as a base model on 8 H-100 GPUs and monitored performance in MLflow
