Swasti Sahay

AI/ML Lead with 10 years of experience designing and delivering enterprise-grade Generative AI and Agentic systems. Currently leading architecture and development of multi-agent LLM platforms integrating LangGraph, LangChain, and hybrid Text-to-SQL/Text-to-NoSQL pipelines across MongoDB and PostgreSQL. Strong expertise in building scalable RAG systems, LLM guardrails, contextual memory frameworks (Redis), and Azure-based AI ecosystems for real-time enterprise intelligence and intelligent automation.

• Certificate of Appreciation for Outstanding AI Performance in IoTDE and Connected Services.
• Multiple Beyond Performance Awards and On-the-Spot Excellence Awards for innovative AI contributions.

• Applied Data Science Bootcamp, Massachusetts Institute of Technology (MIT-ADSB), 2020.
• Developing AI Applications using Python and TensorFlow, University of Oxford (Online), 2020.
• IP Filed: Gen-AI-based Contextual Response Generation from Multimedia Streams.

• Multi-agent Maritime Analytics Assistant (KAI Agent), Built a multi-agent maritime analytics assistant that answers complex operational and financial queries by orchestrating MongoDB and Postgres in real time.
• Agentic RAG Chatbot for Safety Compliance, Designed and implemented an Agentic AI-powered chatbot using LangChain, Azure AI Search, and MongoDB Atlas Vector Search, enabling context-aware, multimodal question answering from OSHA and ANSI safety documents.
• Real-time Hazard Detection in Drone Imagery, Built a VLM-based AI pipeline using GPT-4o to detect safety hazards from drone images and video streams.
• Gen AI Powered Safety Tracker, Worked on design and development of Gen AI powered Safety Tracker, an inventive safety monitoring system that cross-references visual data against company policies.
• Document Intelligence for Safety Certificates, Developed an OCR and LLM-based solution using Azure Cognitive Services and Regex to automate text extraction, validation, and hazard risk categorization from scanned safety certificates.
• Task Hazard Identification & Mitigation System, Created a RAG-powered hazard analysis engine where users input a task description and receive OSHA-compliant hazard identification and mitigation measures.
• LLM Evaluation & MLOps with Azure Prompt Flow, Designed and implemented Prompt Flow pipelines for evaluating LLM outputs, improving reliability and accuracy of hazard detection responses.
• AI Architecture for Workplace Safety Automation, Designed a scalable, Azure-hosted AI ecosystem integrating Custom Vision, Prompt Flow, and RAG-based reasoning for predictive hazard identification.

• Generative AI & Large Language Models:

GPT-4o, Phi-3.5, Retrieval-Augmented Generation (RAG) Pipelines, Azure Prompt Flow, Vision Language Models (VLMs), Embedding Models, Prompt Engineering, Fine-Tuning (LoRA)

• Agentic AI & Orchestration:

LangChain, LangGraph, Semantic Kernel, Redis

• Deep Learning & Computer Vision:

Convolutional Neural Networks (CNNs), Object Detection (YOLOv5, YOLOv8), Segment Anything Model (SAM), Image Processing, OpenCV, TensorFlow

• Machine Learning & Data Science:

Regression Models, K-Means Clustering, Principal Component Analysis (PCA), Feature Engineering, Model Optimization

• Programming & Data Handling:

Python (Seaborn, Plotly, Matplotlib), SQL, Data Cleaning, Data Preprocessing, Data Annotation

• Cloud Platforms & AI Services:

Microsoft Azure (App Service, AI Studio, Custom Vision, OCR, Azure AI Search, Azure OpenAI Studio, Azure Prompt Flow), Google Cloud Platform (GCP)

• Databases & Vector Search:

MongoDB, PostgreSQL, DuckDB, MotherDuck, Vector Search (MongoDB Atlas)

• DevOps & Development Tools:

Podman, Docker, Jupyter Notebook, VS Code

• Architecture & Integration:

API Integration, Document Intelligence, AI Architecture Design, Proposal Writing