AC.

Aayush Chaudhary

AI Engineer

Focused on interpretability and safety. Building AI systems that are transparent, understandable, and aligned with human values.

View Project Get in Touch

About

High-agency AI Engineer driven by continuous learning, specializing in AI research with a focus on interpretability and safety. Proactively stays current with emerging technologies while bridging technical expertise and creative problem-solving across AI, Linux, and game development domains.

Education

Savitribai Phule Pune University

BE in Artificial Intelligence & Data Science

2021 – 2025

Technical Skills

Python PyTorch Langchain FastAPI Docker GCP Linux C++ SQL

Experience

AI Engineer

Altrd – Mumbai

Jul 2024 – Present

  • Engineered AI backend for meeting transcription, improving speaker identification by 40%
  • Improved personalized product recommendation systems by 85%
  • Developed Gen AI applications with NLP, LLM fine-tuning, and RAG systems
  • Implemented solutions using Flask/FastAPI with Docker on GCP

Research Intern

AI Institute of South Carolina

Sept 2024 – Present

  • Researched attack techniques for state-of-the-art text-to-image models
  • Working on AI Safety and Alignment under Prof. Amitava Das

Projects

CavaAthleisure AI Shopping Assistant

Agentic conversational AI system that provides personalized fitness apparel recommendations

CavaBot
Hi Sarah! Based on your marathon training and previous purchases, I'd recommend our Ultra Flex leggings in sage. They have enhanced compression for long runs and match your color preferences.
Do those have a pocket for my phone?
Yes! The Ultra Flex leggings have a zippered side pocket that fits most smartphones. Should I add size 6 to your cart? Your last 3 leggings purchases were this size.

Developed an advanced conversational RAG chatbot for CavaAthleisure that provides hyper-personalized shopping recommendations based on customer data. The system intelligently integrates multiple data streams including user fit preferences, body type, workout habits, color preferences, and purchase history to deliver contextually relevant product suggestions in a natural conversation flow.

Key Technical Features:

  • • Multi-vector RAG system with memory persistence for contextual conversations
  • • Agentic behavior with tool-calling capabilities to access inventory, customer profiles, and order history
  • • Fine-tuned LLM for athletic apparel domain knowledge and sizing consistency
  • • Advanced embedding model for semantic matching between customer preferences and product attributes
  • • Real-time inventory integration with purchase completion capabilities
Python LangChain FastAPI Vector DB LLM Fine-tuning Redis

Results:

  • • 63% increase in conversion rate for recommended products
  • • 47% reduction in product returns due to better fit recommendations
  • • 4.8/5 customer satisfaction rating with the AI assistant
Case Study

Attention Visualization (Attn1)

Interactive tool for visualizing attention mechanisms in large language models

Interactive visualization of token attention in Llama 3.1

Attn1 provides insights into how language models process and attend to different parts of input text, making the internal mechanisms of LLMs more interpretable. This tool helps researchers understand attention patterns, contributing to AI interpretability research.

Python Flask Hugging Face D3.js
View Project

Achievements

GATE DA

AIR 3867 (2025)

WorldQuant Championship

Bronze level, Global rank 257

StableCode Hackathon

9th rank out of hundreds of teams

Open Source

Contributions to Yabai, Amethyst, Waterfox

Contact

Interested in AI interpretability, safety, or want to discuss potential collaborations?

aaayush.dev@gmail.com