Shubham Yadav

- Developed a large production ready AI-pipeline for object tracking for an MNC - Building data pipeline for large scale qualitative analysis for an NGO - Building an LMM powered, Accessibility focused SaaS product for a startup; providing fundraising support

Datacluster Labs crowdsourced vision datasets for companies to build their AI products. - Reached 100k users with 0 marketing, biggest Vision data crowdsourcing platform built in India - Clients included Samsung, Atos, etc.

Worked as a Deep Learning engineer and Software Developer - Developed and improved upon Vision models that - Worked on the development of multiple production pipelines at Uncanny - Took initiative on cleaning of ~1 TB of image data, resulted into a 10% mAP across classes - Worked on merging two major production pipelines

Research aim: Early detection of Septic Shock in patients using Time Series Analysis of 4 major vitals Approach: Developed a novel algorithm for multivariate time series prediction. Used Matrix Profile along with Dynamic Time Warping and Random Forest to find patterns from 6-hour long time series of BP, heart rate, respiration rate and SpO2 and classify patients as abnormal or normal. Achieved: 69% AUPRC, 81% accuracy Worked on: R, Python, Tensorflow

Worked on creating an end-to-end data pipeline and machine learning model for forecasting sales of pallets at Brambles (World's largest distributor of reusable pallets and containers) Accomplished: Developed a deployable pipeline that reduced the forecasting error below 1% in the estimated demand of pallets in the next 2 months. Worked on: Python, Tensorflow, Pytorch