Ankur Mahiwal
Lead Data Scientist @Oliver Wyman
Signup · Get unlimited contacts
WORK HISTORY
Lead Data Scientist @Oliver Wyman
AI Retail Attribution — LLM Inference at Scale- Engineered batch-processing system leveraging Claude LLMs on Databricks SQL for distributed inference over millions of product attributes- Built Node.js/Express middleware orchestrating batch submission, real-time job tracking, and result retrieval against Databricks compute- Implemented structured data extraction with Pydantic BaseModel schemas for type-safe, deterministic JSON outputs- Integrated multi-modal inputs (images, descriptions, web-scraped content) with configurable LLM parameters (temperature, tool choice, completion windows)- Designed confidence-scoring layer annotating outputs with evidence levels and extraction rationale for downstream QALLM + RAG Document Extraction Platform- Proposed and led an LLM+RAG solution extracting structured fields from unstructured property valuation reports, replacing a manual, error-prone process- Shipped interactive web UI for non-technical users to test extractions, view side-by-side PDF comparisons, and validate outputs in real time- Automated pipeline processing thousands of documents, cutting turnaround time and improving consistency- Extended platform to agricultural-land valuations, unlocking incremental client revenue LLM Evaluation Framework — Financial QA- Led full-cycle evaluation of GPT-4 variants on FinQA benchmark using role-based prompting (solver, financial agent, programming aide)- Built flexible evaluation harness — prompt templates, automated scoring, step-by-step reasoning extraction — reproducible across teams- Delivered recommendations to leadership on prompt strategy and model selection for enterprise LLM use
EDUCATION
Indian Institute of Management Tiruchirappalli
Master of Business Administration - MBA, Strategy and Finance
Motilal Nehru National Institute Of Technology
Bachelor of Technology - BTech, Computer Science
SKILLS
ABOUT ANKUR MAHIWAL
I build production AI systems — LLM pipelines, RAG applications, and ML models — that go into use at large financial institutions. Currently leading GenAI initiatives at Oliver Wyman, where I\'ve shipped a distributed LLM inference system on Databricks processing millions of product attributes, an LLM+RAG platform that automated structured extraction from property valuation reports (replacing a manual, error-prone process), and an LLM evaluation harness for financial QA on the FinQA benchmark. The stack: Claude and GPT-4, Pydantic for structured outputs, Node.js/Express middleware, Databricks SQL for distributed inference. Before the AI focus, I built classical ML at scale — credit risk models (XGBoost) for a Malaysian digital bank, climate risk assessment frameworks under NGFS scenarios, and data pipelines in PySpark, Kafka, and Snowflake. Started my career as an SDE at Cisco working on router infrastructure, where I drove unit-test coverage from 15% to 55% and won the internal automation hackathon. Claude Certified Architect. Reach me at m••••••••@gmail.com.
This profile is compiled from publicly available professional sources. Unifers is not affiliated with or endorsed by LinkedIn. Request removal of this profile.