My name is Prakash,
Current - AI Research Scientist at invideo.ai working as Generative AI team lead building multi-model systems from last 1.5 year, Previously
- At qure.ai I worked as
Senior Director - Data scienceleading 3D (CT scan - Chest and Brain) team for 3 years. - At fractal.ai, worked as
Data scientistfor one year and later worked asSenior Data scientistfor 3 years.
You can find me at
I have been working in the deep learning field for over 9 years. I enjoy learning things from scratch and love writing and teaching about them. I am known for my attention to detail—you'll often find me documenting my findings in thorough reports. For any project, I start by gathering information, collecting data, and establishing a solid evaluation framework. I focus on understanding the fundamentals and take a mathematical approach to every problem.
- Talking heads: Multimodal Generative AI system. Trained GAN models on 5k+ hours of audio+video data. experimenting with diffusion foundational models now [2025]
- SyncNet: Contrastive learning, CLIP type model for audio+video alignment. Trained on 10k+ hours of audio+video data. [2024]
- Identifying defects from Product X-rays. Non-destructive testing . self supervision (200k+ images) + segmentation (25k+ images).
- 3D object detection: Finding nodules on Chest CT. Trained on 200k CT Scans. [2023]
- 2D object detection: Identifying Humans in drone footage - thermal and normal cameras - Trained on 100+ hours of data. [2022]
- 2D object detection and long tail classification: Quantifying brand presence in retail shelfs, Detecting 200+ objects from single image and identifying 5000+ SKUs. Trained on 100k+ images [2020-21]
- Image classification: Identifying diabetic retinopathy from Eye Images. Trained on 50k images [2018].
This is more of a research project
Below are some competations where I was in Top-25
- Using focal loss on deep recommender systems - 14th Position
- AV-Fractal-hiring hackathon on Forecasting - 9th Position
The following blogs are written by me on various platforms.
- All my generative AI blogs are written here. details on diffusion, flow, score and GAN based models are detailed.
- some of blogs on foundation models were written here
- Diffusion models
- Diffusion transformers
- Variational AutoEncoders
- VQVAE
- A primer on score based generative modelling
- A primer on flow based generative modelling
- CLIP
- AIM - AutoRegressive Image models
- SimMIM - A simple framework for Masked Image Modelling
- vision transformers
- Understanding Object detection metrics and errors
- ATSS
- Fully convolutional one stage object detection
- Guide to build faster RCNN using PyTorch-2020
- Overview from R-CNN to SSD
- The intution behind retinanet
- Structural reparameterization
- CSPnet
- Classification architecture review Alexnet-SENet
- Understanding Resnet and Resnext Part1 Part2
- Almost any image classification using Pytorch-2018
- Transfer learning using keras - 2017
- Intro to self-supervised learning 2020
- Cutout
- Recurrent Neural networks using tensor math
- Back-Propagation is very simple. Who made it Complicated ?
- 💼 any freelance work? do reach, email :)
- 💬 ask me about anything, i am happy to help;