Sourabh Raj

Architected and scaled a multi-agent AI ecosystem, leading the system design for specialized agents including PipeSpec, Profiling, Schema, SQL, and Genie to automate complex data workflows and enhance large-scale analytical precision. Engineered a unified Model Context Protocol (MCP) to orchestrate and scale a suite of specialized AI agents including PipeSpec, Profiling, Schema, and SQL, streamlining complex system design and enhancing cross-agent data retrieval for the Genie ecosystem.

Leading a 20-member cross-functional team spanning Product Management, AI Research, Development, Frontend, and Data Engineering to deliver scalable AI solutions. Led the architectural design and migration of a large-scale agentic AI infrastructure from AWS to GCP, ensuring scalability, reliability, and cost optimization.

Developed a scalable solution to create a system which can store unlimited number of chat in Amazon DynamoDB. Used vLLMs for efficient entity extraction. Created a ChatGpt-like playground using Amazon Bedrock. Led a team of AI engineers to design and fine-tune healthcare-specific Large Language Models (LLMs), enabling intelligent summarization of clinical notes and device data. Hands-on experience with Databricks as a central data lake, integrating CI/CD pipelines for automated deployment. Designed, coded, and deployed a context-aware AI healthcare assistant for diabetes management, integrating Model Context Protocol (MCP) and autonomous AI Agents. Led and mentored a healthcare-focused team to build a real-time multi-agentic diabetes virtual assistant.

Developed advanced tools for building flexible, multi-modal AI agents powered by Large Language Models (LLMs). Designed and developed full scaled AI Agentic system for Text-to-Sql product. Led a cross-functional team to design, develop, and deploy scalable machine learning solutions. Developed an agentic Retrieval-Augmented Generation (RAG) model using Langraph. Led and mentored a FinTech-focused AI/ML team delivering solutions for risk analytics, fraud detection, and trading optimization. Built domain specific agents and workflows for multi-turn, multi-modal, multi-user conversations with LLM Gen AI driven agents. Developed and implemented similarity models, embedding models, and topic modelling, alongside performing EDA on text data. Experience with FastAPI for building high-performance APIs and integrating LangChain for NLP workflows. Performed entity resolution using Neo4j and developed demand prediction models using Temporal Fusion Transformers (TFT). Worked on customer fake review prediction using LLMs and NLP. Developed and deployed scalable machine learning models and pipelines using Google Vertex AI. Fine-tuned large language models (LLMs) and developed custom language models. Proficient in quantization techniques including LoRA and QLoRA. Implemented data virtualization solutions using Starburst. Deployed accelerators and proof-of-concept solutions on GCP. Developed Python code for anti-money laundering solutions using Neo4j. Developed predictive modeling and anomaly detection models. Developed dynamic pricing strategies for flight fares in the aviation sector.

Utilized advanced querying, visualization and analytics tools to analyze and process complex data sets. Led a team of AI engineers in developing predictive models for patient readmission risk. Coordinated efforts across engineering and compliance teams to ensure all AI models met HIPAA and healthcare regulatory standards. Developed game-theoretic RL simulations with OpenSpiel for vaccine distribution and treatment prioritization. Developed a RAG pipeline for document extraction and query generation using LLMs. Fine-tuned Large Language Models on healthcare datasets using Peft-LORA. Worked on cloud deployment for AWS for model deployment. Worked on MLOps techniques to streamline ML workflows and improve model deployment. Worked on ASMR and SMR model prediction using LightGBM. Designed and implemented NER powered semantic search leveraging LLMs. Built medical domain specific search engine using hybrid search. Built RAG Pipeline for chatbots and conversational AI. Built Oncology related RAG using Gen AI. Proficient in Prompt Engineering and text processing techniques. Worked with Vector Databases like pinecone.

Analyzed complex data and identified anomalies, trends, and risks to provide insights to improve internal controls. Explored use of OpenSpiel in early stage drug discovery by modelling multi-agent interactions. Used OpenSpiel to prototype reinforcement learning agents for supply chain decisions. Communicated data requirements to the global team. Led and mentored a cross-functional team for project delivery. Implemented CNN, Masked CNN, and YOLO models for object detection. Built and trained neural networks to detect emotions from images. Worked on OCR, face detection, and recognition. Worked with Data Visualization team and Power BI.

Built machine learning models in credit risk assessment. Provided proof of concept for machine learning in credit risk modelling. Worked with Data Preprocessing and Feature Engineering. Implemented machine learning algorithms for predictive models in credit risk. Evaluated credit risk models and interpreted results. Worked on time-series analysis and built ARIMA, Exponential Smoothing, and other models for forecasting. Applied time series forecasting to financial forecasting and demand planning. Implemented reinforcement learning framework for algorithmic trading using OpenSpiel.

Developed machine learning model for predicting prices using AWS Sagemaker. Collected and prepared data for training price prediction model. Tested troubleshooting methods and documented resolutions for ML models. Deployed price prediction model on AWS Sagemaker. Adjusted design parameters to boost performance and incorporate new features.