Sai Sekhar Dharmireddy

Generative AI Engineer at Goldman Sachs

Spearheaded production-ready LLM pipelines with GPT-4, BERT, and T5, orchestrated via LangChain, LlamaIndex, and Haystack; achieved over 92% accuracy in legal contract summarization and risk assessment. Fine-tuned LLMs (LLaMA, T5) using RLHF, PEFT, LoRA, and Prompt Engineering; implemented with Hugging Face Transformers, PyTorch, and TensorFlow. Architected multi-agent workflows for regulatory compliance and transaction surveillance. Translated business objectives into use cases, test cases, and labeled datasets. Deployed scalable inference APIs with FastAPI, Docker, Kubernetes, and AWS SageMaker; CI/CD with GitHub Actions and Jenkins. Designed low-latency vector search with FAISS, Pinecone, and Chroma DB for RAG-based QA. Led experimentation, version control, and visualization with MLflow, Weights & Biases, TensorBoard. Communicated results to engineers and stakeholders; collaborated with legal/compliance; embedded explainability with SHAP, LIME, and Captum.

Machine Learning Engineer at CVS Health

Engineered NLP pipelines (spaCy, NLTK, TextBlob) and Transformers to extract structured medical data from EMRs for optimized care delivery. Built multi-label classification models (PyTorch, TensorFlow, Scikit-learn) deployed via Flask and Django for internal dashboards. Developed document QA systems with LangChain, RAG, and Prompt Engineering for real-time care team interactions. Ensured governance, privacy, and bias auditing (SHAP, LIME, Fairlearn). Automated data ingestion/ETL with Airflow, Pandas, NumPy, and AWS Lambda; improved processing speed by 40%. Maintained deployments on S3, EC2, SageMaker; implemented observability with Grafana, CloudWatch, and MLflow.

Data Scientist - NLP & Computer Vision at ACKO General Insurance

Developed multimodal fraud detection models using OpenCV, CLIP, and ResNet; deployed with TensorFlow, ONNX, and TorchServe. Created hybrid models for claims automation by fusing OCR text with image features using Stable Diffusion, DALL·E, and BERT. Built distributed PySpark pipelines on Hadoop/Hive for real-time streams; deployed via Azure ML Studio and Cognitive Services within claims workflows. Accelerated training with CUDA, TensorFlow, Keras, and JAX (mixed-precision). Managed data across MongoDB, Cassandra, and Redshift, supporting feature stores.

Data Scientist - NLP & Forecasting at Bank of America

Designed market sentiment models using BERT, T5, and LSTM for financial forecasting and equity scoring. Built enterprise chatbots with Dialogflow, Flask, and LangChain; integrated risk scoring for internal audits. Implemented reproducibility and multi-environment deployment with MLflow and Docker across GCP and internal servers. Developed scalable ETL workflows with Apache Airflow and Oozie processing over 2 TB of daily data. Created hybrid retrieval pipelines (PostgreSQL, HBase, Hive) for batch and real-time use cases; applied SHAP and LIME for explainability to support regulatory compliance.

Generative AI Engineer at Goldman Sachs

Engineered LLM-based AI pipelines for contract summarization and risk analysis using GPT-4, BERT, and T5; orchestrated with LangChain, LlamaIndex, and Haystack, achieving >92% accuracy. Developed Legal Document Summarization Agent leveraging LangChain, FAISS, and OpenAI API to automate financial due diligence. Fine-tuned foundation models (LLaMA, T5) using RLHF, PEFT, LoRA, and prompt engineering, improving domain-specific summarization accuracy. Designed and managed Agentic AI workflows with Crew AI, LangGraph, and Autogen for cross-agent coordination in compliance and knowledge extraction. Integrated Model Context Protocol (MCP) for secure, standardized context sharing across multiple AI agents and enterprise APIs, reducing redundancy by 30%. Built context-driven tool-using agents that dynamically interact with internal APIs, databases, and documentation repositories to automate financial report validation. Deployed scalable inference APIs via FastAPI, containerized with Docker, orch

Machine Learning Engineer at CVS Health

Engineered advanced NLP pipelines using spaCy, NLTK, TextBlob, and Transformers to extract structured medical data from EMRs, optimizing care delivery. Built and deployed multi-label classification models with PyTorch, TensorFlow, and Scikit-learn, integrated into Flask/Django dashboards for provider teams. Applied transfer learning and fine-tuning on BERT and T5 for domain-specific NLP tasks, improving model accuracy and convergence speed. Developed document QA and semantic retrieval systems using LangChain, RAG, FAISS, and Pinecone, enabling real-time care team insights. Automated monitoring and retraining workflows with Grafana, CloudWatch, and Airflow, ensuring model reliability and minimizing drift. Built data ingestion and preprocessing pipelines using Pandas, NumPy, and AWS Lambda, improving throughput and reducing manual intervention by 40%. Ensured fairness, explainability, and compliance using SHAP, LIME, and Fairlearn, collaborating with risk teams for audit readiness.

Data Scientist - NLP, Computer Vision & Predictive Analytics at ACKO General Insurance

Engineered multimodal fraud detection models combining image and text data using OpenCV, CLIP, and ResNet, deployed via TensorFlow, ONNX, and TorchServe, reducing fraudulent claims. Developed hybrid OCR-image models leveraging Stable Diffusion, DALL·E, and BERT for claims automation, improving processing accuracy and efficiency. Built distributed PySpark pipelines on Hadoop and Hive to process real-time insurance data streams, accelerating ETL throughput by 40%. Deployed ML solutions using Azure ML Studio and Cognitive Services, integrating them into claims decisioning workflows for faster approvals. Optimized deep learning training using CUDA, TensorFlow, Keras, and JAX, implementing mixed-precision training to reduce computation time and costs. Managed structured and unstructured data across MongoDB, Cassandra, and Redshift, supporting scalable feature engineering and model input pipelines. Implemented model monitoring and performance evaluation pipelines, ensuring reliability, comp