Hi, my name is

Abhijeet Anand.

I solve problems using machine learning.

I'm a Data Scientist specializing in machine learning, deep learning, and large-scale system design. Currently, I'm building ML-powered solutions for dynamic pricing, personalization, and predictive analytics at Zepto.

About Me

Hello! My name is Abhijeet and I'm passionate about leveraging data science and machine learning to solve complex real-world problems. I graduated from IIT Roorkee in 2024 with a B.Tech in Mechanical Engineering, though my heart has always been in the world of data and AI.

Fast-forward to today, and I've had the privilege of working at Zepto, where I was fast-tracked to Data Scientist II in just 9 months. I've built everything from reinforcement learning systems for dynamic pricing to deep learning models for query segmentation processing 25M+ queries. Previously, I worked at Timely.AI and FilterPixel, building ML solutions for image processing and automated editing.

I'm passionate about building scalable ML systems that drive real business impact. Whether it's using Bayesian methods for price optimization or implementing multi-armed bandits for dynamic pricing, I love finding elegant solutions to challenging problems.

Here are some technologies I've been working with:

  • Python
  • PySpark
  • SQL
  • Machine Learning
  • Deep Learning
  • AWS
  • Databricks
  • Kafka
  • Streamlit
  • Docker
  • Neural Networks
  • Reinforcement Learning
Headshot

Where I’ve Worked

Data Scientist II @ Zepto

June 2024 - Present

  • Designed and deployed a Reinforcement Learning-based Multi-Armed Bandit framework for dynamic city × SKU pricing, delivering a 1.7% OPD lift and 1.2 Rs GPPO gain
  • Developed a Bayesian Hierarchical Model to estimate SuperSaver price elasticity, leading to a 0.5 Rs GPPO increase through experiment-validated pricing
  • Fine-tuned a GLiNER-based NER model for real-time query segmentation, achieving 89% accuracy and extending from 3.5M cached queries to real-time processing of 25M+ queries
  • Built a tree-based User × Category Premiumness Model to classify users into cohorts, improving targeting of high-value first-time buyers for recommendation ranking
  • Derived Store × Category Premiumness signals from user-level cohorts, enabling data-driven SKU onboarding and reducing premium SKU overstocking

Some Things I’ve Built

Other Noteworthy Projects

view the archive
  • Folder
    GitHubExternal Link

    Eye Desktop - AI Photo Search System

    Eye Desktop is a production-ready AI-powered semantic photo search application that enables users to find images using natural language queries. Unlike traditional keyword-based search, Eye understands the semantic meaning of queries and returns visually relevant results.

    Built with Google's SigLIP-512 (state-of-the-art vision-language model), achieving 78% top-1 accuracy and 93% top-10 accuracy. The system processes 2.4 images/second on CPU and delivers search results in under 100ms. Features a FastAPI backend, React/TypeScript frontend with Tauri desktop framework, FAISS vector similarity search, and SQLite for metadata storage.

    Key achievements include optimizing FAISS from L2 distance to cosine similarity (40% better discrimination), resolving macOS segmentation faults through import order fixes, and implementing efficient batch processing with memory management for 85% faster indexing.

    • Python
    • PyTorch
    • SigLIP-512
    • FastAPI
    • React
    • TypeScript
    • FAISS
    • Tauri
    • SQLite

What’s Next?

Get In Touch

Although I’m not currently looking for any new opportunities, my inbox is always open. Whether you have a question or just want to say hi, I’ll try my best to get back to you!