I'm a Senior Data Scientist at Octus, where I specialize in building LLM-powered AI and agentic systems that transform financial technology operations. With 7+ years of experience in data science, I excel at the intersection of AI research and applied product development.
My expertise lies in designing and deploying high-performance AI solutions, with a sharp focus on document intelligence, knowledge retrieval, and autonomous agents. I thrive in fast-paced, execution-driven environments, delivering production-ready solutions within 1-3 month development cycles.
Beyond the technical aspects, I believe in AI systems that ship, scale, and solve real business problems. My approach is hands-on, execution-driven, and deeply focused on building AI that works in the real world.
Download ResumeCurrently building a multi-agent system using AWS Bedrock to analyze credit agreements autonomously. This system will streamline document processing, reduce manual review time by 80%, and improve accuracy in financial document analysis.
Currently developing an end-to-end system that extracts, chunks, and summarizes key insights from financial term sheets using AWS Textract and Bedrock. The system uses RAG to identify critical terms like covenants, interest rates, and maturity schedules—dramatically speeding up diligence and improving interpretability.
Building an iOS app to manage a personal library using barcode scanning and the Google Books API. The app lets users scan, organize, and search their collection, complete with smart shelf placement and QR codes for tracking. Designed with Swift, inspired by BookBuddy, and tailored for customization and offline-first usability.
Large Language Models (LLM)
Generative AI
Natural Language Processing (NLP)
Document Intelligence
RAG Systems
Machine Learning
AWS Services
AWS Bedrock
AWS Textract
Data Engineering
ETL Pipelines
Python & SQL
Multi-Agent Systems
Chain-of-Thought Prompting
Autonomous AI Agents
Information Retrieval
Context-Aware Systems
Agent OrchestrationJuly 2023 - Present
• Architect and implement document classification and summarization systems using AWS Bedrock Agent
• Design and deploy scalable document processing pipelines leveraging AWS Textract
• Develop asynchronous AI workflows for enhanced processing efficiency
• Enhance RAG frameworks for improved retrieval accuracy and context-aware insights
• Lead cross-functional collaboration to deliver production-ready AI solutions within 1-3 month cycles
November 2022 - April 2023
• Engineered ELT pipeline for migrating 100K+ records from BigQuery to PostgreSQL
• Developed XGBoost regression models for weekly performance metrics across 800+ radio stations
• Implemented hierarchical clustering algorithms for analyzing spin metrics of 50K+ songs
• Collaborated with engineering teams on code review, testing, and model deployment
June 2021 - October 2022
• Led Data Science team in strategic planning with C-suite executives
• Built market segmentation models for $800M end customer and OEM market
• Optimized e-commerce strategies resulting in 300% engagement and 150% sales increase
• Developed ARIMA-based forecasting models for 80,000+ products
• Implemented pricing optimization solutions with $25M quantifiable effects
• Created 25+ PowerBI dashboards saving $150K in resource allocation
August 2020 - April 2021
• Analyzed 500K+ political ads from 2018 Congress and 2020 Presidential elections
• Applied NLP methods to identify patterns in targeted political advertising
• Investigated economics of online fundraising campaigns
September 2019 - January 2021
• Developed centralized information portal for medical student mentorship
• Built ETL pipelines integrating REST APIs and Blob storage
May 2020 - April 2021
• Implemented Airflow and PySpark workflows for unstructured text analysis
• Developed statistical models improving quotation hit rates by 20%
• Built Snowflake data pipelines processing 100K transactions
• Created anomaly detection models with 80% accuracy
August 2016 - July 2019
• Architected NoSQL database schema for airline crew-scheduling applications
• Managed 2TB distributed NoSQL database across three environments
• Developed Python automation scripts for Data Quality Management
• Built PySpark and Kafka framework for real-time flight operations data
Interdisciplinary Data Science
Metallurgy and Material Science Engineering
Built a scalable document processing pipeline using AWS Textract and Bedrock that improved document classification accuracy by 40% and reduced processing time by 60%.
Developed an autonomous system of specialized AI agents that analyze financial documents, extract key information, and generate insights with 95% accuracy.
Designed and implemented an improved RAG system that increased retrieval accuracy by 35% and reduced hallucination rates by 50% in financial document analysis.
akshaypunwatkar@gmail.com
New York, NY
