Prathamesh Palkar

Generative AI Solutions & Agentic Workflows (Jan 2023 – Present)

Architected a multi-agent Intelligent Digital Assistant using Python, LangGraph, LangChain, and Azure AI Foundry along with frontier language models like GPT 5.1 and Claude 4.6 to automate enterprise tax workflows, cutting manual data entry by 80%.

Architected and deployed containerized GenAI APIs to AKS, ensuring high availability and secure secret management via Key Vault.

Built Retrieval-Augmented Generation (RAG) systems with OpenAI APIs and Azure AI Document Intelligence for conversational interaction with financial PDFs, optimizing performance using hybrid search, re-ranking, and metadata filtering to increase retrieval precision by 20%.

Integrated LLM observability metrics to monitor relevance, context precision, and hallucination rates using OPIK observability framework.

Designed a voice-to-voice conversational interface using OpenAI Realtime APIs for seamless natural-language interaction.

Developed a Generic Data Extractor using GPT-4o and Anthropic Sonnet models to parse diverse document types, standardizing data across workflows.


Machine Learning & Computer Vision (XTRACT K1 ML) (Jun 2020 – Present)

Engineered and trained computer vision pipelines for document intelligence, performing extensive image annotation and training YOLOv7 models to successfully detect and extract complex document structures including headers, footers, paragraphs, and tables from PDF images.

Managed distributed training, monitoring, and scalable deployment of ML systems utilizing Azure Machine Learning Studio compute clusters.

Designed ML microservices in Python & Flask achieving 95% classification accuracy in identifying tax forms.

Deployed Named Entity Recognition (Flair + heuristics) for precise field extraction and integrated APIs with .NET microservices for seamless enterprise deployment.


ESG Benchmarking & Predictive Pipelines (Oct 2022 – Oct 2023)

Developed a highly scalable data pipeline in Azure Databricks to process large-scale SEC Edgar 10-K filings, leveraging PySpark and distributed processing for rigorous NLP and ESG topic modeling.

Integrated complex ML prediction results into Power BI dashboards for real-time ESG trend analysis, enabling rapid business intelligence decisions.


LLM Metrics & Reporting (Jan 2024 – Present)

Built a Power BI dashboard tracking tokens per minute utilization (TPM) and PTU across AI systems, optimizing compute costs and context window utilization.

Reduced LLM provisioning overhead by identifying redundant token usage and optimizing model invocation.

SalesEdge (C# ASP.NET, MS SQL Server, WCF)

• Built a web application for a major U.S. tobacco manufacturer to manage retail payments, promotions, and employee assignments.
• Created a bulk payment approval module for executives, cutting manual effort, and streamlining approval workflows.
• Optimized code and SQL queries to reduce screen render time from 4 minutes to 30 seconds, significantly boosting performance.