Senior Consultant 1 at Ernst & Young Global Delivery Services (2025-10 – Present)
- Developed an agentic AI system to automate adaptation of PSD-based marketing creatives across multiple aspect ratios, engineering a LangGraph-based pipeline leveraging Vertex AI (Gemini Pro) and Google Cloud Vision API to combine LLM reasoning with deterministic image analysis while preserving design consistency and enforcing creative constraints.
- Developed intelligent Smart Object-based image cropping and title treatment pipelines, enabling image repositioning and resizing within Photoshop PSDs while preserving creative intent through safe-zone enforcement, relative title positioning, face/subject avoidance, and composition-aware layout optimization.
- Implemented regeneration workflow with human-in-the-loop, incorporating designer feedback, leveraging LLM to classify crop and title-related feedback and trigger targeted creative regeneration for iterative refinement.
- Built and deployed Python microservices on Google Cloud Run, exposing REST APIs for Adobe Experience Manager (AEM) while persisting workflow in PostgreSQL and storing assets in Google Cloud Storage (GCS).
- Collaborated with Java teams responsible for AEM UI integration while owning end-to-end AI orchestration, prompt engineering and cloud deployment.
- Reduced creative asset turnaround time by ~80% and manual design effort by ~60%, enabling designers to focus on creative decision-making instead of repetitive asset adaptation.
Senior Consultant 1 at Ernst & Young Global Delivery Services (2025-10 – Present)
- Developed and deployed an agentic AI system on AWS using LangGraph, automating direct response and commercial ad instructions for multiple agencies.
- Built an email-triggered workflow that initiates agent execution, integrating AWS Lambda, SQS, and SNS for event-driven processing and notifications.
- Implemented LLM-powered agents using Amazon Bedrock, with FastAPI and MCP servers exposing tool endpoints for structured agent actions.
- Designed human-in-the-loop checkpoints and custom persistence using Amazon S3 and PostgreSQL, deployed on ECS, to support long-running, auditable workflows.
- Implemented LLM observability and logging using Phoenix, enabling traceability, debugging, and performance analysis of agent interactions.
- Delivered the solution successfully to production with 2 weeks of hypercare support, achieving a stable release with no rollbacks, improving operational accuracy by 80% and significantly reduced manual processing effort through automation.
Senior Data Science Engineer at Ksolves India (2025-06 – 2025-10)
- Engineered a GenAI-powered agentic framework using LangGraph, Chainlit, Python, OpenAI API, and LangChain to automate case handling workflows.
- Implemented custom LLM chains for intent detection, entity extraction, and retrieval-augmented generation (RAG) using FAISS vector search for knowledge base integration.
- Utilized PostgreSQL for conversation and workflow state persistence.
- Designed an end-to-end pipeline with automated ingestion of case data from CRM systems, followed by LLM-driven summarization, human feedback-based query generation, log analysis for root cause identification, and automated resolution and knowledge base content generation.
- Containerized the framework using Docker with environment-specific configuration for staging and production environments.
- Achieved a 43% reduction in case resolution time through intelligent triaging, automated documentation, and real-time backend integration.
Consultant at AbsolutData (acquired by Infogain India Pvt Ltd.) (2022-03 – 2025-06)
- Developed a Retrieval-Augmented Generation (RAG) system using LangGraph, FAISS, and LLMs to address physician queries with high accuracy.
- Enhanced retrieval and contextual understanding from patient clinical notes, improving relevance and reducing hallucinations in responses.
- Implemented vector database search (FAISS) for efficient similarity search over large-scale clinical data.
- Designed a pipeline to integrate LLM-powered query answering with patient information.
- Conducted A/B testing and shopper analysis using python (regression models) for in-store marketing campaigns and key performance indicators (KPIs) such as planogram effectiveness, portfolio performance, and share of shelf, for multiple geographies across the globe.
- Performed data cleaning, exploratory data analysis (EDA), model data preparation, and execution.
- Developed insights and narratives from model outputs using Python and Excel to support data-driven decision-making.
- Executed code within a pre-configured Docker environment and contributed to the migration of the workflow to Databricks, leveraging Azure Blob Storage.
- Engineered a simulator to recommend budget allocations for various subcategories of tactics across different marketing channels (e.g., TV, Digital – Facebook, YouTube), to achieve a given target volume using scipy's optimize and differential evolution.
- Calculated performance metrics such as volume, revenue, and return on investment to assess the effectiveness of marketing strategies. Incorporated the additional impact of brand communication and Halo effect from different brands into the optimization process.
- Implemented an extra layer of optimization on commercial levers (trade, price, distribution, etc.) in cases where the desired volume was not achieved through advertising tactics.
- Developed an Interactive Python API on Streamlit to generate S-curve parameters.
- Collaborated with backend and frontend teams to upgrade and integrate the simulator, contributing to the development of a user-friendly interface, and with the Data Science Lead to educate the team to provide a detailed understanding of the code base.
- Supported in the project to prepare an automated tool to extract density of tea shops for selected markets.
- Extracted data for multiple geographies from the Google Places API.
- Cleaned and processed the data to extract count of tea shops based on defined conditions specific to different zip codes.
- Visualized the outputs on the map for a better view and simplified reading.
- Automated & deep learning-driven solution to calculate planogram compliance by comparing the planogram of previous and after images clicked by sales representatives leveraging Python, Azure, Computer Vision (YoloV5, InceptionResNetV2).
- Worked to fix the deviations in the results of the existing model due to variations in image data.
- Developed a new model tailored for identical objectives but with a distinct data category from data labeling for input to the model, deep learning pipeline setup, hyperparameter tuning to achieve optimal results to mapping outputs on the respective images for planogram visualization.
- Codified the process to merge multiple log files and flag the compliances for the advertisements being played in movie theatres.
- Aimed to enhance the efficiency of marketing resource allocation at a product category level (e.g., Beverages, Snacks, Beauty, etc.).
- The primary goal was to analyze and optimize the distribution of resources across diverse platforms, ensuring the maximization of return on investment.
Analytics Intern at HighRadius Corporation (2021-01 – 2022-03)
- Worked in the Analytics Department to extract and automate data for reports and interactive dashboards using MySQL, Snowflake, Python on different business products, which include Collections, Deductions, Credit.
