Hi, I am

Atharva Chandak.

I build intelligent systems.

I'm an ML Engineer currently building multimodal AI solutions at eBay. My journey spans research at Harvard, Mila (UdeM), CMU, and IISc - from continual learning and 3D segmentation to robotic vision - alongside industry experience in Deep Learning and scalable Full-Stack engineering.

Explore My Work
Atharva Chandak

Experience

ML Engineer 2

eBay
Sept 2025 - Present
  • Building Multi-Modal models to match user listings to catalog products for 1B+ items.
  • Finetuning SOTA VLLMs for fine-grained image data extraction in Sports Trading Cards.
  • Prototyped a buyer experience assistant for hyper-personalized recommendations with agentic memory.
PythonPyTorchLangChain

Software Engineer

Wells Fargo
Jul 2023 - Sept 2025
  • Full-stack developer for high-volume PL credit card/loan platforms processing 1M+ monthly transactions.
  • Prototyped a scalable LLM chatbot for merchants, reducing application time by 30%.
  • Built backend API microservices for POS EMIs and a SaaS customer insights portal.
JavaReactSpring BootPython

Research Assistant

Mila / UdeM
Jan 2023 - Jun 2023
  • Researched Continual Object Detection for autonomous driving, cutting catastrophic forgetting by 73%.
  • Created on-device incremental OD methods for robotics using vision transformers.
PyTorchHugging Face

Research Intern

Airlab, CMU
Aug 2021 - Dec 2021
  • Engineered the vision pipeline for TartanAviation, generating 3.1M high-resolution images and 700k+ auto-labeled aircraft annotations.
  • Designed optimized models for real-time aerial vehicle detection and tracking on drone hardware with multi-camera input.
  • Analyzed OOD climatic conditions to improve model robustness across diverse weather scenarios.
PythonPyTorchOpenCV

Summer Intern

Wells Fargo
May 2022 - Jul 2022
  • Built an end-to-end AI speech chatbot for home lending queries, integrating speech-to-text and TTS APIs for a full voice interaction loop.
  • Implemented NLP pipeline for intent and entity extraction, achieving F1 scores of 93.4% and 83.5% respectively.
NLPRasaPython

Visiting Research Intern

Visual Computing Group, Harvard
Jan 2022 - May 2022
  • Contributed to PyTorch Connectomics, adding cellpose model support for neuron instance segmentation.
  • Explored semi-supervised methods and designed an affinity learning + transformer pipeline to improve 3D segmentation accuracy.
PyTorch3D SegmentationTransformers

Research Intern

AIRL, IISc
Jun 2021 - Dec 2021
  • Worked on Generalized Continual Zero-Shot Learning, integrating incremental learning with ZSL for realistic DL deployment.
  • Extended the setup to task-free, out-of-distribution generalization across diverse Computer Vision tasks.
Zero-Shot LearningContinual LearningPyTorch

Summer Intern

CSIR-CEERI Chennai
May 2021 - Aug 2021
  • Applied traditional CV algorithms (FAST, ORB, BRISK) with ML classifiers (SVMs, KNNs) for texture classification of industrial images.
  • Implemented segmentation networks to detect cracks and faults in leather, enabling automated quality inspection.
OpenCVScikit-LearnSegmentation

Education

BITS Pilani Logo

BITS Pilani

Bachelor of Engineering in Computer Science

Coursework: Data Structures, Algorithms, Cloud Computing, Machine Learning, Deep Learning, Computer Vision, Natural Language Processing.

Featured Work

AdQuestA

AdQuestA: VQA for Advertisements

A multimodal VQA framework for ads achieving 58.58% accuracy on a new 94k+ QA pair benchmark. Integrates RAG with LLaVA for knowledge-grounded ad understanding. Published at WACV 2025.

VLMRAGLLaVA
LIFT-Net

LIFT-Net

A dual-pathway transformer attending to raw RGB and illumination-corrected frames for light-invariant action recognition. Achieved SOTA on HMDB51, UCF101, ARID, and InfAR.

TransformersAction RecognitionMulti-modal
Drone Delivery

Drone for Parcel Delivery

Optimized UAV navigation, control systems, and QR-based detection in a simulated environment, improving delivery success rates and accuracy. Top 5 in E-Yantra Competition.

ROSOpenCVControl Systems
Image Super Resolution

Image Super Resolution

A GAN based model to convert low resolution images to high resolution. Read and implemented various Deep Learning algorithms like SRGAN & ESRGAN.

PyTorchGANsComputer Vision

Technical Arsenal

Languages

Python JavaScript TypeScript Java C++ HTML/CSS

ML / AI

PyTorch 🤗 Hugging Face 🦜 LangChain OpenCV Scikit-Learn TensorFlow NumPy Pandas

Frameworks

React NextJS Node.js Spring Boot Flask Django

Cloud & Tools

Docker AWS Azure Git Jenkins Kafka MongoDB MySQL

Major Achievements

  • Top 30 in India in IMC Prosperity 2, a global algorithmic trading competition by IMC Trading.
  • 3rd Place in AI RoboSoccer Competition (148 teams) using PPO-based RL agents.
  • Ranked 598 & 2449 out of 1.23M in JEE Main & Advanced.
  • Top 5 in IIT Bombay's E-Yantra Robotics Competition.