I’m a Lead Machine Learning Engineer at LevelAI, with four years of hands-on experience in NLP, LLMs, and GenAI. My journey into AI wasn’t conventional—I don’t have a formal degree in AI or Computer Science. Instead, I’ve taught myself everything, from the depths of machine learning to its broadest applications.

To build a strong foundation, I’ve taken rigorous courses like :
1) Gilbert Strang’s Linear Algebra
2) Stanford CS109 Probability for Computer Scientists by Chris Piech
3) Stanfdord CS231n Convolutional Neural Networks
4) CS224n Natural Language Processing with Deep Learning by Chris Manning and team
5) CS229 on Machine Learning by Andrew Ng.
I’m currently exploring reinforcement learning through Berkeley’s Deep RL course

I stay at the forefront of the field by keeping up with the latest research, from Flash Attention and ROPE to RLHF and the pioneering work by OpenAI. On the applied side, I’ve brought theory to life—implementing models from scratch, writing extensive PyTorch and Python code, and turning research papers into real-world solutions.

As a Lead Machine Learning Engineer at LevelAI, I spearhead innovative projects that leverage advanced AI and NLP techniques to transform and enhance customer service experiences. My role involves developing, fine-tuning, and implementing models that improve the efficiency and effectiveness of contact center operations. Here are some of the key projects and contributions I've made at LevelAI:
1. LLM Based Autoscoring of Contact Center Agents
2. Contextual Emotion Detection in Dialogue Systems
3. RAG based Chatbot for solving complex customer queries and helping contact center agents serve their customer better
4. Automatic Customer Satisfaction and Customer effort detection from contact center conversations

Patents and Innovations
Patent Granted in US Patent Agency for Development of a dynamic intent detection system detecting intents from customer-agent interaction in real time . This system is adaptable to real-time client modifications, surpassing traditional static models.

Publications
GWNET: Detecting Gravitational Waves using Hierarchical and Residual Learning based 1D CNNs
- This was part of our gold Winning solution for Kaggle Comeptition on Detection of Gravitational Waves
- We built an end-to-end 1D Convolutional Neural Network architecture that can be applied directly on raw time-series data collected by these interferometers to detect gravitational waves. The key model aspects are ability to capture hierarchical features using parallel Convolutional blocks and residual learning via custom Residual layers.

In my free time I love to participate in Kaggle competition to further enhance my skills in AI . Some profound achievements include
- Winner of Kaggle Days Championship Regionals, representing India in the finals in Barcelona, Spain.
- Ranked as a Kaggle Competitions Master, with 3 gold and 10 silver medals, and participation in over 30 competitions.
- Ranked as a Kaggle Notebooks Grandmaster, with 16 gold medals, showcasing solutions to various problems across different verticals.