Achyut Ha P

Led the end-to-end development of enterprise GenAI applications on AWS (SageMaker, EMR, Redshift) and Bedrock, designing multi-agent runtimes with MCP tool-calling and CrewAI orchestration for role-based workflows. Built Retrieval-Augmented Generation pipelines combining BM25 + vector search with FAISS and Pinecone, enabling grounded financial document analysis and knowledge retrieval. Implemented domain-adapted fine-tuning with PEFT (LoRA/QLoRA) to improve summarization and risk entity extraction. Created secure FastAPI services with OAuth2 and RBAC; deployed on EKS for scalable, real-time inference. Built and evaluated LLM agents using LangChain and LangGraph, automated ingestion/validation of SEC filings and research reports, achieving ~33% workflow efficiency gains. Integrated Neo4j knowledge graphs to enhance fraud detection and compliance analytics. Established monitoring with CloudWatch, X-Ray, and Prometheus; introduced responsible AI guardrails (SHAP/LIME). Conducted automated

Led production-grade ML models for patient risk prediction using XGBoost, Random Forest, and deep learning on Google Cloud Vertex AI; delivered end-to-end pipelines with Vertex AI, MLflow, and Cloud Build to accelerate model release cycles by ~25%. Engineered scalable data ingestion and analytics with Cloud Dataflow, Cloud Storage, and BigQuery; integrated external social determinants to improve risk stratification. Implemented OCR and document intelligence with Cloud Document AI; developed NLP pipelines (BERT, spaCy, Transformers) for clinical notes; deployed low-latency inference endpoints on Vertex AI. Applied SHAP/LIME explainability to support HIPAA-compliant audits and clinician trust; built model monitoring and drift dashboards with Looker Studio and Cloud Monitoring. Containerized services with Docker and GKE; delivered GenAI-driven clinical documentation workflows using MedLM (Med-PaLM 2) on Google Cloud for ambient note generation.

Developed forecasting, pricing, and demand-planning ML systems for retail datasets; built scalable ML pipelines in Python and PySpark with Azure data services; engineered time-series features via SQL/dbt and Snowflake. Implemented ARIMA, Prophet, LSTM, and deep learning models, achieving ~25% forecast accuracy improvement for high-volume SKUs. Built ensemble models (XGBoost, Random Forest) for pricing and promotions; incorporated external signals (Nielsen, promotional calendars, social sentiment). Operationalized batch and scheduled inference via Azure Functions with CI/CD across 1,800+ locations; established governance via Azure Purview and telemetry via Azure Monitor. Delivered ML-driven forecasting dashboards in Power BI/Tableau; collaborated with merchandising and supply-chain teams to translate forecasts into pricing/inventory decisions; version-controlled pipelines with Git and Azure DevOps.