Spandan Maaheshwari

Pioneered a multi-agent GenAI system using LangChain, LangGraph, and OpenWebUI with custom router agents to classify user intents and route tasks to GPT-4 powered analytics modules, expanding automated insights across enterprise data by 80%. Streamlined text-to-SQL conversational agent pipelines with automated schema normalization, anomaly detection, and Pinecone-backed RAG context injection, cutting data integration errors by 90% and accelerating dashboard delivery by 40%. Fine-tuned LLMs (T5, Mistral, Claude) to improve protein-sequence classification accuracy by 32%, built scalable ETL pipelines in PySpark and AWS Glue to process 3 TB+ genomic datasets with HIPAA/HL7 compliance; created explainable AI dashboards (SHAP, LIME) and deployed containerized services with Docker/Kubernetes; sustained >99.9% availability.

Established a scalable RAG pipeline using Azure Data Factory, ADLS Gen2, Unity Catalog, and Databricks, automating ~10k document ingestions into Delta Lake and improving embedding similarity by 81% via LLaMA-2 70b. Deployed a real-time chatbot REST API using Streamlit, Kubernetes-based Lakehouse monitoring, leveraging Azure DevOps CI/CD with 120+ unit tests and MLflow to evaluate toxicity, readability, and minimize model hallucinations.

Established a scalable RAG pipeline using Azure Data Factory, ADLS Gen2, Unity Catalog, and Databricks, automating ~10k document ingestions into Delta Lake and improving embedding similarity by 81% via LLaMA-2 70b. Deployed a real-time chatbot REST API using Streamlit, with Kubernetes-based Lakehouse monitoring and Azure DevOps CI/CD with 120+ unit tests; evaluated toxicity and readability to minimize model hallucinations using MLflow.