Work Experience

Freelance AI Engineer, - Dublin, Ireland Apr 2025 - Present

• Worked with the Founders of Healiom, Neurosonic, VizCo, Getty Images, (Format Techologies)[https://useformat.ai/].
• Actively working on building and scaling AI Agent Workflows for Format Technologies.
• Benchmarked Different Embedding models on internal medical datasets of Healiom.
• Implemented and Benchmarked different ML models and introduced AutoML for stenosis prediction for Neurosonic.
• Compared the search results for different queries using Google Search vs GPT Search, developed a RAG pipeline to simulate and compare search results for Generative Engine Optimization.
• Developed a Multi-Agent System to convert Gherkin Style BDD specifications into Cypress UI tests at Getty Images.

Senior AI Engineer — NLP / LLMs / Multimodal
LinkedIn, Dublin, Ireland — Aug 2022 to Nov 2024

• Developed language and multimodal deep learning models for feed, search, and notifications, contributing to +0.47% revenue and +1.2% engagement.
• Built and evaluated LLM-based personalization features.
• Collaborated cross-functionally with infra and product teams for production deployment.

Data & Applied Scientist | Software Engineer II — ML/NLP
Microsoft, Redmond & Atlanta — Dec 2020 to Jul 2022

• Developed and scaled a backend API for a Forecasting Service now supported in Microsoft Dynamics and Excel.
• Developed a Generative SLM (T5) for aspect and reason based sentiment analysis on customer support feedback forms.
• Introduced CI/CD practices into the team for auto-deployment of services.

AI & Deep Learning Engineer — NLP / Data Engineering
Decooda, Atlanta, USA — Jul 2019 to Nov 2020

• Built data pipelines using Matillion + Snowflake and applied ML/NLP to extract business insights from unstructured customer feedback.
• Developed a Graph RAG based FAQ chatbot

AI Smart City Research Assistant — Computer Vision
Georgia Tech, Atlanta, USA — Jan 2019 to May 2019

• Built CV models to detect and track vehicles in road work zones.
• Estimated vehicle counts, speeds, and flows using deep learning models.

Machine Learning Intern
NASA Ames Research Center, California — May 2018 to Aug 2018

• Applied super-resolution with Deep CNNs on low-res climate data to predict the CO2 levels in California.
• Benchmarked multiple DL architectures for geospatial tasks.

Teaching Assistant — Knowledge-Based AI (Head TA)
Georgia Tech, USA — Jan 2018 to Dec 2018

• Created exam material for 800+ students in the OMSCS program.
• Worked on Jill Watson, the class’s AI teaching assistant.

Software Development Intern
Citrix R&D, Bangalore — Jan 2017 to Jun 2017

• Automated API creation based on in-house config files.

Computer Vision Intern
Wipro Technologies, Bangalore — Jun 2016 to Aug 2016

• Developed lane and pedestrian detection models using OpenCV and custom algorithms.

Accelerators

Founders Talent Accelerator
Dublin, Ireland - Jan 2025 - Apr 2025

● Participated in a Talent Accelerator to find co-founders and create a startup. Explored areas in AI For SportsTech, Agentic Workflows for Task Mining, ML based Insights Platform for Longitudinal Gut Microbiome Analysis.

Side Projects

Content Creation

1. Actively post about AI developments, libraries and papers on LinkedIn. Grew to 13k+ followers. Checkout some of my best posts on my profile page.

2. Blog about my projects and ideas on Medium.

Open Source Contributions

Core Contributor To TRL An active core contributor to TRL the go-to open source library for most pre-training, mid-trainig and post-training algorithms like SFT, PPO, GRPO, DPO etc.

Liger-Kernels Contributions
Open-source contributions to RLHF kernel functions, including SimPO and CPO. Also wrote Triton kernels (e.g., GroupNorm) to accelerate training.

Klarity Contributions
Benchmaked how hallucinations can be detected using entropy metrics. Enforced structured outputs in JudgeLLMs and optimized static embeddings to stabilize token similarity results.

AI Agents

MCP Server to Order Coffee Explored how to create custom MCP servers by playing around with DevTerminalShop which lets users order coffee using an API or the terminal.

Agent-to-Agent Protocol Demo Show cased the different features/functionalities of the A2A protocol proposed by Google to standardize the communication with agents hosted anywhere.

• Showed how model cards work and where they are hosted.
• Highlighted how clients can subscribe to the updates of an Agent.

Gut Microbiome Research Bot Explored SmolAgents and created an Agent that uses Pubmed/Biorxiv to identify research papers and answer questions that help a practitioner keep up-to date with the world of Gut Microbiome research.

Company Reasearch Assistant Explored LangGraph andd created an agent that automates the process of crawling company websites, extracting product and service information, classifying the company, and generating reports—including Google Docs—using AI and various integrations.

AI Modeling and Training

Create your own GRPO Trainer A tutorial to implement your own GRPO Training Loop without using veRL, TRL etc. Also goes into the weeds of the math behind the GRPO objective.

Benchmarking SOTA LLMs on Medical Definition Recognition

Mechanistic Interpretability and Superposition 101 A 101 on how deep learning models can learn more features than dimensions through the concept of superposition.

Exploring Sink Tokens
Researched how Transformer models (encoder and decoder) consistently direct attention to a small set of “sink” tokens like [CLS] or the first token. Found that dense self-attention may be unnecessary, suggesting more efficient LLM training strategies.

• Wrote Medium articles summarizing findings and experiments.

SemEval 2020 — Humor Detection
Finished Top 5 in Task 7 of SemEval 2020. Built Siamese Transformer models to assess the funniness of edited news headlines.

• Published paper: LMML at SemEval-2020 Task 7

Block Stacking with RL
Designed a 2D block stacking environment using Q-learning. Studied the influence of oracles and demonstrations on suboptimal Q-tables.

Kaggle NLP Competitions (x4 Top Finishes)
Top 3 finishes in 4 out of 5 university-led NLP tasks, including POS tagging, co-reference resolution, and dependency parsing.

Education

MS in Computer Science (Interactive Intelligence)
Georgia Institute of Technology, Atlanta, USA — 2019

• GPA: 3.91/4.00

B.Tech in Computer Science
Manipal Institute of Technology, India — 2017

• GPA: 9.29/10 (Top 10% of class)

Publications

• LMML at SemEval-2020 Task 7 — Rated humor in edited news headlines using Siamese Transformers, SemEval 2020.
• Question Answering with EOL — Fall AAAI Symposium 2018
• Optic Disc Approximation — IJERT, 2016

Skills

• Programming: Python, Java, Scala, C#, SQL
• Machine Learning & Deep Learning: AI Agents, PyTorch, TensorFlow, Transformers, RLHF, Triton, Sklearn, Computer Vision
• Data Engineering: Snowflake, Matillion, Spark, Ray
• Tools: Git, Docker, Azure, Jupyter, VSCode, AWS
• Research Areas: NLP, LLMs, Attention Mechanisms, Reinforcement Learning, Hallucination Detection
• Libraries/Frameworks: Pytorch, Pytorch-Lightning, TRL, Transformers, EmbedAnything, Sentence Transformers, Langchain, LLamaIndex, SmolAgents, Crew.ai, LangGraph, LiteLLM, Sklearn, Tensorflow, Keras, Pinecone, Together.ai — Connect with me on LinkedIn.