Akshay Ballal

MACHINE LEARNING ENTHUSIAST

Hello

Here's who I am and what I do

Welcome to my personal website! I am an AI developer pursuing a Master’s in Artificial Intelligence at TU Eindhoven, where I am constantly applying First Principles thinking to solve complex problems. I am the creator of EmbedAnything, a Rust-based library designed to optimize embedding pipelines, making them faster and more reliable. As the head of the Machine Learning team at Team Serpentine, I am leading the development of AI solutions for Brain-Computer Interfaces. With a passion for advancing technology, I am committed to using AI for social good and exploring green technology innovations. Let’s collaborate to shape a better future through cutting-edge solutions to real-world challenges.

Recent Blogs

12 Jan 2025

Optimize VLM Tokens with EmbedAnything and ColPali

ColPali is a late-interaction vision rag model that enables text search within images. It can locate specific pages in PDFs containing relevant text, even when the text exists as part of an image. This powerful capability forms the foundation of Vision RAG systems.

29 Sept 2024

Build ReAct Agents using SLMs from Scratch

Build a function-calling agent using a custom Unsloth model with a LoRA adapter, switching between reasoning and function execution, following ReAct-style reasoning.

6 Aug 2024

REINFORCE Algorithm

In this article we will look at the first policy gradient algorithm: REINFORCE

Featured Projects

EmbedAnything

0 stars 0 forks

A Rust-based library designed to optimize embedding pipelines, making them faster and more reliable. Features include:

Local, ONNX, and cloud embedding model support
Multimodal processing (text, images, audio)
Vector streaming for memory-efficient indexing
91.2% Rust, 8.5% Python codebase

View on GitHub View on PyPI

Contributor to Candle

Contributing to Candle, a minimalist ML framework for Rust with 16k+ stars. Added three key models to enhance the framework's capabilities:

ColPali - Vision RAG Model
Splade Model - Sparse Bert Model for BM25
Reranker Model - Cross-encoder Architecture

View Contributions