Data Science Researcher | Multi-Agent Systems Specialist | LLM Application Developer | RAG Pipeline Expert

Vinit Sutar

Specializing in building and deploying scalable LLM applications and delivering high-impact business solutions through multi-agent systems.

Philosophy

"Architecture is about the trade-offs we choose, not just the tools we use.".

Experience

Senior Chief Manager - Data Science

Shriram Finance Limited | Bangalore

  • Led a team of 5 Data Scientists for Propensity Models, Delinquency Modelling, SMS Engine, LLM Finetuning, FD Cohorts, Market Mix Modelling for campaign optimization.
  • Improved Propensity Models which increased the conversions across Fixed Deposits products by INR 520 Crores since March 2025.
  • Implemented a Market Mix Modelling best practices by optimizing the marketing campaigns to save spends by 5 pct.
  • Engineered a high-scale RAG pipeline on Azure Databricks, deploying a vector search endpoint with HNSW indexing.
  • Enabled Tool Calling capabilities with Gemma3 for precise information retrieval.

Associate Consultant - Data Science

Conneqt Digital - Client - Anheuser-Busch InBev | Bangalore

  • Developed conversion attribution model using GLMNET and statsmodel Logistic Regression for South Africa Region with accuracy of 98 pct.
  • Implemented a conversion package to calculate incremental lift with model error between +-5 pct.
  • Implemented 2 power (n-1) fractional factorial design which halved the time and resources to run Randomized Control Trials.
  • Developed Total Distribution approach with CUPED backend to help understand the impact of portfolio recommendations over n-4 months.
  • Developed propensity score matching for using Dynamic Time Warping, Locality Sensitive Hashing and Faiss+Nearest Neighbour Approach for 200k stores.

Data Scientist

Impact Analytics | Bangalore

  • Modernised prophet pipeline to get trend from various hierarchies to increased the accuracy by 3 pct of a regression model[cite: 5].
  • Defined and developed an imputation logic which improved imputation from 0.3 pct to 0.7 pct compared to existing approach[cite: 6].
  • Designed a new pipeline for store clustering using Dynamic Time Warping which takes in account units sold with time[cite: 8].
  • Mentored a team of two in transitioning and knowledge transfer about client and process flows[cite: 7].

Data Scientist

HiMedia Laboratories | Mumbai

  • Micro-organism Detection using OpenCV where detection accuracy was improved by 40 percent compared to existing algorithm[cite: 12].
  • Developed an annotation pipeline using Raspberry Pi, integrated with CMOS camera which saved INR 80/per image for 500 images[cite: 13].
  • Implemented an Active Learning approach using Mask RCNN to reduce annotation efforts, saving INR 6,00,000[cite: 14].
  • Developed a detection pipeline using Mask RCNN to detect, count and calculate growth rate of Micro-organisms[cite: 15].
  • Saved instrument cost of INR 1,00,000 and increased accuracy from 80 pct to 94 pct compared to existing hardware[cite: 16].

Professional Development & Intern

AAIC Technologies | Speedtail Ventures

  • Built 15 Portfolio Projects on Data Science ranging from Predictive Modelling to Recommender Systems[cite: 19].
  • Utilized Tesseract OCR, Regex, and PowerBI for data processing and visualization.

POCs to working Product