Aman Dalmia

I love learning and building things 👷🏽

2023 -
I returned to HyperVerge after 5 years. Currently, I lead AI for HyperStart, a legal co-pilot and we are building SensAI, a socratic AI tutor, for our non-profit arm HyperVerge Academy where we support students from low-income backgrounds to get jobs in the tech sector.

You can join in too:
  • SensAI is open-source, fork it/clone it/play around with it
  • Join our community of builders in AI + education and share how you would like to contribute
  • Look at our public roadmap and share your ideas/requests
  • Help spread the word and gather more feedback on how we can do better
[I'm hiring Senior ML Engineers for my team. If you are one or know someone who might be interested, please consider applying or passing it on to them.]
2022 - 2023
Joined an early-stage startup named WisdomCircle as a Consultant Tech Lead. Built the first version of product with Joanna and the rest of the team to help retirees find meaningful things to do post-retirement and set up the engineering team.
2020 - 2022
I worked as a Software Engineer at Avanti Fellows, a non-profit founded in 2010 by Akshay Saxena that works with students from low-income backgrounds to support them in getting into good engineering and medical colleges, providing a pathway out of engineering and medical colleges, providing a pathway out of poverty. We built several open-source products to improve how our students learn. Notably, we built Plio, an open-source tool that converts any YouTube video into an interactive lesson plan and provides educators all the learning data at the level of each student. It increased the completion rates on our videos by 5x and was used by > 40000 students in Haryana during the COVID-19 pandemic. We were part of the 7th Cohort of The Nudge Incubator and won the Tools Competition by Schmidt Futures under the "Accelerating Learning" track in 2022-23. Plio continues to be used by Avanti Fellows to support students in their learning and is being used by other non-profits like The Apprentice Project (TAP) in their programs as well.

I built the first version of Avanti's Quizzing Engine, an open-source platform for students to practice and prepare for their exams. The quizzing engine is in active development and regularly being used by thousands of students. Here is a demo of the quizzing engine in action!

In the journey, we got to go to incredible places and meet amazing people. We even gave a presentation only using memes. Visiting Tehri for a retreat with other tech-focused non-profits is a memory that I will always cherish. You can read about our experience here!
2018 - 2020
I was the founding team member of Wadhwani AI, a non-profit dedicated to using AI for social good. We built AI-based solutions for pest-management in cotton farms (featured on the PyTorch blog, won the Google AI impact challenge for $2M, published a paper in KDD), estimating the weight of newborns from a video to identify low-birth weight babies (accepted as best paper at Computer Vision for Physiological Measurement (CVPM) Workshop, CVPR 2024) and screening for COVID-19 using cough sounds (paper published at ICLR 2021 AI for Public Health and Machine Learning for Preventing and Combating Pandemics Workshops, code).
2014 - 2018
B.Tech at IIT Guwahati with a major in Electronics & Communication Engineering. But I hardly focused on my major until my curriculum included a lot of math and ML. I got into programming in my first year [android development], worked with a lot of startups on campus, started (and failed) my own startup in education, and then, got into Machine Learning, mostly learning on my own, making open-source contributions to scikit-learn, working with Prof. Amit Sethi on applying deep learning to predict cancer recurrence using histopathology images, working with Prof. Ashish Anand during my Bachelor Thesis Project on predicting splice sites using genome sequences (paper) and interning at Niramai [detecting breast cancer in thermal images using deep learning (paper)] and HyperVerge [change detection in geospatial images using deep learning].
pet projects
Genie GPT Small web-app to upload your documents and give a shareable link to anyone to chat with your documents. Very crappy. Built in the early days of the release of GPT4. No longer maintained.
CoreML Built a generic end-to-end ML framework as an abstraction to quickly run experiments. This was done right at the time when PyTorch lightning was released and I didn't pursue it further.
Deep Learning Book: Chapter Summaries (> 1k on Github) was my attempt along with Ameya to make the Deep Learning Book easier to understand by providing more visualizations for several concepts along with code samples where possible.
Notes for the Reinforcement Learning course by David Silver along with implementation of various algorithms. (> 700 on Github).
PyTorch implementation of Sinusodial Representation networks (SIREN) paper (> 250 on Github).
.
WannaPark: A real-time car parking (model) using deep learning in a 2-week hackathon during my summer internship at Qualcomm in 2017. We used deep learning to match the combo of number plates + the driver's face to ensure a car was not being stolen as well as object detection on CCTV camera footage to automatically identify empty parking spots.
publications
NurtureNet: A Multi-task Video-based Approach for Newborn Anthropometry
CVPM Workshop, CVPR 2024
Yash Khandelwal, Mayur Arvind, Sriram Kumar, Ashish Gupta, Sachin Kumar Danisetty, Piyush Bagad, Anish Madan, Mayank Lunayach, Aditya Annavajjala, Abhishek Maiti, Sansiddh Jain, Aman Dalmia, Namrata Deka, Jerome White, Jigar Doshi, Angjoo Kanazawa, Rahul Panicker, Alpan Raval, Srinivas Rana, Makarand Tapaswi
Impact of data-splits on generalization: Identifying COVID-19 from cough and context
ICLR 2021 (workshop)
Makkunda Sharma, Nikhil Shenoy, Jigar Doshi, Piyush Bagad, Aman Dalmia, Parag Bhamare, Amrita Mahale, Saurabh Rane, Neeraj Agrawal, Rahul Panicker
Pest management in cotton farms: an AI-system case study from the global south
2020
Aman Dalmia, Jerome White, Ankit Chaurasia, Vishal Agarwal, Rajesh Jain, Dhruvin Vora, Balasaheb Dhame, Raghu Dharmaraju, Rahul Panicker
Exploring deep learning networks for tumour segmentation in infrared images Captioning
2020 (Journal)
Siva Teja Kakileti, Aman Dalmia, Geetha Manjunath
Using the Chou’s 5-steps rule to predict splice junctions with interpretable bidirectional long short-term memory networks
2020 (Journal)
Aparajita Dutta, Aman Dalmia, R Athul, Kusum Kumari Singh, Ashish Anand
highlights