Atharva

I’m Atharva, a Machine Learning Engineer focused on LLMs, RAG and AI agents. I like building things that ship, writing what I learn, and competing on Kaggle for fun. This site collects my notes, projects, and occasional essays. See my about page to read about my past work in much detail. Also check out my blog where I write about what I learn occasionally. The best way to follow my activity is X (Twitter).

💼 Work with me

I am open to consulting work and collaborations. Please reach out to me at atharvaaingle@gmail.com

📰 News

Participated in WSDM Cup - Multilingual Chatbot Arena on Kaggle: Earned a solo silver medal placing 31st out of 950 teams. Challenge was to develop a reward model (used in RLHF stage) for multilingual human conversations on the chatbot arena (formerly LMSYS). Read details about my solution here.

What I am doing lately

I am diving deep into distributed learning
I am exploring AI agents and looking where they can be genuinely used.

🏢 Who I’m working with currently

Wolters Kluwer (Jan 2023 - Present): Machine Learning Engineer:

I mostly work on Document AI for legal and financial workflows. I’ve shipped multiple LLM/RAG services to production and am currently exploring where agents are genuinely useful in our products.
Recent work that I’m proud of: Successfully instruction tuned a Qwen-2.5-VL model to extract structured information from billions of scanned images of very complex and different formats and deployed the complete system to production. This project involved preparing training data from scratch, establishing evals, fine-tuning the model, and implementing robust production serving infrastructure.

Weights & Biases (May 2022 - Present): Ambassador:

Writing technical reports and educational content for the community, turning new research and tools into simple, working examples and teach the ML community how to use them and demonstrate best practices with W&B platform.
Engineering optimized Kaggle notebooks with added support for W&B tracking and monitoring tools.

📮 Blog

Recent posts and notes. You can subscribe to my RSS Feed here:

Date	Title
8/8/25	MUVERA - Google’s FDE approach for speeding up multi-vector retrieval
4/10/25	Using Google’s Agent Development Kit and Agent2Agent
4/7/25	Deep reinforcement learning: Integrating neural networks with RL
4/5/25	Supervised learning vs deep learning vs reinforcement learning
1/27/25	LLMs are machine learning classifiers
10/5/24	RAG techniques: From naive to advanced
5/1/24	AI and regulatory compliance in finance
4/29/24	What is model risk management in finance?
4/25/23	Building a Q&A Bot for Weights & Biases
2/24/23	BLIP-2: A new Visual Language Model by Salesforce
1/24/23	Is PyTorch 2.0 Faster Than PyTorch 1.13?
11/29/22	SetFit: Efficient Few-Shot Learning Without Prompts
11/16/22	DeepMind Flamingo: A Visual Language Model for Few-Shot Learning
10/15/22	Hugging Face Accelerate Super Charged With Weights & Biases
8/16/22	3D Segmentation with MONAI and PyTorch Supercharged by Weights & Biases