Hey, I'm Atharv!
I'm a Machine Learning Engineer who's been building AI for 3+ years. I love creating things that actually help people.
I'm really into NLP and Computer Vision, and I'm always looking for a good challenge to tackle next.
Projects
Grammatical Error Correction Model with Dynamic Context Learning
Used BERT model to learn optimal splits within the paragraph/document ensuring maximal contextual information is preserved in the resulting chunks. Achieved an improvement of 77%, 19.61%, and 10.49% F0.5 score on CoNLL-2014, BEA-Dev, and FCE-Test datasets respectively from current SOTA models. Paper accepted at BDA 2023.
Glycosylated Hemoglobin Prediction
Introduced an optimized weighted ensemble model that predicts the risk of Type 2 Diabetes. Worked on improving the performance by developing custom cost function for weights. Technologies: Scikit-learn, Seaborn, Ensemble Learning, SciPy.
AI Image Caption Generator
Developed an Encoder-Decoder model using CNN and LSTM to generate captions from an image. Encoded Images using InceptionV3 model and words using GloVe. Technologies: Keras, CNN, LSTM, GloVe, Transfer learning.
Smart India Hackathon Winner 2022
Won Smart India Hackathon 2022 against the problem statement KK1182 (Shared usage of workshops/labs across india) held at IIT Guwahati.
Experience
Founding Machine Learning Engineer
- Do ML end-to-end — data → training → deployment → monitoring
- Trained a ModernBERT-based ML parser hitting ~97% micro-precision on mapping user queries → Elasticsearch parameters
- Built a GPT-4o powered review analysis pipeline (fetch, clean, summarize, insights on product pages)
- Built AI smart search + product recommendations with a multi-agent setup (Elasticsearch + semantic search + Cohere rerank + RRF + LangGraph)
- Basically turning messy user text into clean signals and making search actually smart
Machine Learning Engineer
- Cut inference time in half (2X faster!) by optimizing the entire model pipeline
- Built a multi-region shoe segmentation model using HRNet and a custom decoder to handle the tricky bits
Earlier experience 6 roles · 2021–2023
Applied Scientist Intern
- Dug into data to find what actually makes new customers stick around for the long haul
- Built predictive models and used Reinforcement Learning to figure out the best way to spend the incentive budget
- Got my research published and presented at the Amazon Machine Learning Conference (AMLC) — pretty cool!
Machine Learning Intern
- Used Diffusion models to generate synthetic "anomalies" so our detection models could learn faster
- Played around with thermal image pose estimation and SAM to push the limits of object segmentation
Deep Learning Intern
- Built a global identity verification system using DBNet to spot the right documents every time
- Slashed latency by 147% and cut GPU usage in half through some serious pipeline tuning
Computer Vision Intern
- Worked on the messy world of food pics — building segmentation and watermark removal for Zomato and Swiggy
- Brought models like U-Net and CLIP to life, wrapped them in APIs, and deployed them on Nvidia Triton
NLP Research Intern
- Turned research papers into real talk by working on open-domain conversation models
- Contributed to Meta's BlenderBot 2.0 (open source for the win!)
- Finetuned and shipped GPT-J-6B on SageMaker using Hugging Face
Research Intern
- Dropped prediction errors by 18% just by tweaking a custom cost function for an ensemble model
- Researched better ways to predict diabetes using data-driven weighted ensembles
Wins & Accomplishments
Runable X Together 2026 Winner
Built and launched a startup from idea to MVP in under 10 hours. Won $1,000 cash and an opportunity for $1M funding via SwarmSpace by Together Fund.
VibeHack 2025 Winner
Took home 1st place among 120 teams with SynthScene — a project reimagining brand advertising through virtual product placement.
Metadome Emerging Tech Hackathon Winner 2024
Won the Metadome.ai Emerging Tech Hackathon, pushing the boundaries of what's possible in immersive spatial computing.
Smart India Hackathon Winner 2022
Won Smart India Hackathon 2022 against the problem statement KK1182 (Shared usage of workshops/labs across india) held at IIT Guwahati.
Amazon ML Summer School 2022
Among the top 3 students from ADGITM who made it to Amazon ML Summer School 2022.
MLH's Neighborhood Hacks Winner
Placed 1st among 300+ participants in MLH's Neighborhood Hacks with AasPaas.
Coders vs COVID Hackathon Winner
Secured 1st among 600+ participants in Coders vs COVID Hackathon with ArchSearch.
HackISOLATION & Innerve Hackathon
Winners of HackISOLATION by IOSD and 2nd runner-up in Innerve Hackathon, by IGDTUW.
Life
Positions of Responsibility
- Machine Learning Lead, Google Developer Student Clubs: Organized workshops on Machine Learning, Supervised and Mentored team of 20+ members
- Founder, Younged: Founded an Ed-Tech platform to teach young aspirants Programming, Designed an efficient and beginner-friendly curriculum and Supervised Teachers
- Technical Lead, TutorateIndia: Managing technical execution to ensure adherence to budget, schedule, and scope. Designed, built, and maintaining a fully Functional LMS Website for the Startup
Away from the keyboard
- I love playing badminton! If you're around the Bangalore HSR area and want to play, hit me up-I'd love to connect.
- Lately, I've been trying to hit the gym as much as I can to stay active.
- You'll also find me playing Valorant (diamond rank no dereanked to plat) or PUBG PC when I'm looking to unwind.
Guestbook
Say hi, share a thought, or just leave your name.