Harshkumar Devmurari

I like to train deep neural nets and play with datasets 🧠🤖💥

harshdevmurari007@gmail.com
While I receive a lot of emails, I'm always open to discussing research and projects. Feel free to reach out, and I'll do my best to respond promptly.

2024 -
coming soon 🧑‍🍳
2023 - 2024
I conducted research in the field of computer vision and it was focused on integration of state-of-the-art computer vision algorithms with path planning to develop a system for autonomous vehicle, presented paper in DJ-ICACTA-23 and published it in IEEExplore journal.

Also at VCET-Solecthon we won ESVC3000+, SEVC2023 championships.
2022 - 2023
I was Perception (Computer Vision) lead at VCET-Solecthon, focused on building robust perception mechanism where objective was to drive a car on road (or any drivable space) without the need of explicit lane markings. Researched and worked on many computer vision tasks including in-house data labelling, improving object detection, dataset curation, lane detection, semantic segmentation, SLAM, etc.

Won SEVC2022 championship and also secured second rank at OpenCV-Core competition

Simultaneously, I took CS50AI course by Harvard University, worked on many projects on AI domain. Explored DL and CNN via course by Andrew Ng.
2021 - 2022
Took CS50X by Harvard University, worked on many projects throughout the course and built and deployed Dronacharya webapp.

Along the way I joined an autonomous-solar-electric vehicle team at college named VCET-Solecthon as autonomus member, at that time objective was to drive a car in constrained environment.
Also worked in aeronautics & aerospace team Airnova as R&D member on drone building and gesture controlled interface for it.
2020 - 2021
Admission in BTech at the University of Mumbai, Vidyavardhini's college of engineering and technology with a major in computer science and a honours in artificial intelligence and machine learning.
Explored the computer science domain and took various courses on web dev. This is where I first got intoduced to deep learning by my brother.
pet projects
autopilot is a research project focused on pure camera based perception mechanisms for autonomous vehicles which dives deep into single-frame, multi-class, state-of-the-art object detection integrating with monocular-cam depth estimation, enabling navigation through dynamic, multi-agent, unstructured environments. Research paper can be seen here and friendly blog post here.
visioblend is a research project focused on noval architecure development for denoising latent diffusion probabilistic model which takes inputs human-drawn sketch(with or without colour information) and converts it into realistic depection with a goal of improving current state of LDDPM. Currently its tested on cats generation but can be generalized to any data. Research paper can be seen here.
dronacharya is a collge recommendation system for engineering admission based on entrace examinations jee and mht-cet. It recommends colleges based on marks(both jee and cet), branch/stem and location integrated with official-data released by jee and mht-cet regarding cutoff score for colleges. Deployed and running up at here. Also refer this
gesture-x is a computer vision based interface to interact with your system without need of mouse. Project was initially inspired for making own cnn-architecture and then was integrated with real-time hand gesture classification and recognition, supporting 20 gestures with over 98% accuracy.
takshak is a farmer assistant project developed with intent to predict and improve agriculture using different ML and DL models. It had features like crop recommendation, yield prediction, disease prediction and weed detection.
misc: I built a lot of other random projects over time.

Many of those stuff was for courses I took like CS50X, CS50AI, deep learning specialization, freecodecamp, etc. Nim playing agent(reinforcement-learning), optimal tic-tac-toe using minimax, shopping-churn predictor, heridity-trait likelihood given simulation, minesweeper-agent knowledge based, CNN architecture for traffic sign, wikipedia based chatbot, IMDB based degree between actors, stock-exchange webapp, and many more....

Some projects are developed from scratch or updated some existing work. 3d-photogrammetry(pifuhd), 3d reconstruction, lane detection, goal-point setting algo in local-map, 300 hours of web-dev, etc.
featured writing
publications
IEEE 2023
Harshkumar Devmurari, Gautham Kuckian, Prajjwal Vishwakarma
arXiv 2024
Harshkumar Devmurari, Gautham Kuckian, Prajjwal Vishwakarma

Also on Google Scholar
misc unsorted

Education

B.Tech in Computer Science Engineering from Mumbai University, 2020-2024

Honours in Artificial Intelligence and Machine Learning from Mumbai University, 2021-2024

Deep Learning Specialization, 2022-2024

Experience

Full-time Engineer at Zeus Learning, specializing in AI/ML applications, Researching and Working on Document Layout Analysis, Multimodal Neural Networks, State-of-the-Art Transformer Architectures, RAG-based Chatbots, NLP from scratch, 2024 - Present

Team Lead in Computer Vision and Autonomous Vehicles research in VCET Solecthon tech team, 2022-2024

Member of Research & Development team in Airnova

Responsibilities included research, modification, and application of data science and data analytics prototypes. Created and constructed methods and plans for machine learning, employing test findings for statistical analysis and model improvement. Trained and retrained ML systems and models, improved current ML frameworks and libraries, and created machine learning applications in accordance with customer needs.

Coordinated with development teams to understand application needs, utilized Python to create scalable code, performed application testing and bug fixing, and integrated storage methods for data. Designed and implemented high-performance, low-latency applications while working closely with front-end programmers.

Projects

Skills

Additional Information