Mayank Bumb

Co-Founder

Boston, Massachusetts, United States6 yrs 8 mos experience

AI EnabledAI ML Practitioner

Key Highlights

Reduced rider impersonation fraud from 15% to 1%
Scaled privacy-preserving pipelines to ~10M frames/month
Developed real-time fraud detection systems handling ~2000 RPS

Stackforce AI infers this person is a Machine Learning Engineer with expertise in AI systems for healthcare and fraud detection.

Contact

Skills

Core Skills

Artificial Intelligence (ai)Machine LearningNatural Language Processing (nlp)Data AnalysisStart-up VenturesLeadershipManagementSoftware DevelopmentApi Development

Other Skills

QAT/PTQoperator rewritesprivacy-preserving infrastructuredata curationfeature extractionmodel trainingon-edge deploymentLarge Language Models (LLM)Cluster AnalysisK-means clusteringSQLTensorFlowSiamese networkIsolation ForestSpeech Recognition

About

I build production-grade AI systems that operate reliably at scale — from research to edge deployment. Currently an Applied Scientist at Cherish Health, working on radar and multi-sensor ML for in-home safety. I’ve productionized on-device inference on Qualcomm hardware (QAT/PTQ, operator rewrites), deployed fall detection systems with 95% recall, and scaled privacy-preserving ground-truth pipelines to ~10M frames/month. Previously at Zomato, I reduced rider impersonation fraud from 15% to 1% (impacting 200K+ riders/day) and shipped real-time fraud detection systems handling ~2000 RPS. Interested in edge AI, multi-sensor perception, scalable ML infrastructure, and LLM-based systems that ship.

Experience

6 yrs 8 mos

Total Experience

1 yr 4 mos

Average Tenure

1 yr

Current Experience

Cherish

2 roles

Applied Scientist

Promoted

Dec 2025 – Present · 6 mos · Boston, Massachusetts, United States · On-site

Productionized on-device inference on Qualcomm by applying QAT/PTQ and rewriting unsupported ops
(Conv3D→stacked Conv2D) for hardware compatibility, preserving ~90% accuracy under device constraints.
Developed scalable, privacy-preserving vision ground-truth infrastructure to generate LiDAR-aligned 3D detection labels from multi-view RGB reconstruction and pose, scaling in-home capture 100× to ~10M frames/month.

QAT/PTQoperator rewritesprivacy-preserving infrastructureArtificial Intelligence (AI)Machine Learning

Machine Learning Engineer

Jun 2025 – Dec 2025 · 6 mos · Boston, Massachusetts, United States · On-site

Owned the production radar Fall-Detection system end-to-end from data curation and feature extraction to model training, on-edge deployment, and drift/performance dashboards, enabling first 100-user device rollout with 95% recall.
Re-architected radar-based people detection into a custom Conical-UNet to better model polar/spherical data structure and improve spatial consistency in sparse, noisy returns boosting recall by 12%.
Built scalable data pipelines and leading internal/external teams to deliver high-quality datasets and save ~$50K/month.

data curationfeature extractionmodel trainingon-edge deploymentArtificial Intelligence (AI)Machine Learning

Adobe

Graduate Student Researcher

Jan 2025 – Jun 2025 · 5 mos · Amherst, Massachusetts, United States · Remote

Published and presented an approach for LLM time-series forecasting method that encodes temporal structure via overlapping patch segments and optional nearest-neighbor augmentation for richer context.
Achieved ~19% lower MAE on Weather and Traffic datasets vs. the S²IP-LLM baseline across different horizons.

University of massachusetts amherst

2 roles

Graduate Teaching Assistant

Sep 2024 – Dec 2024 · 3 mos · Amherst, Massachusetts, United States

I am a grader in the CS 685: Neural Networks course taught by Professor Subhransu Maji and Chuang Gan. For more details, check the following website: https://cvl-umass.github.io/compsci682-fall-2024/

Graduate Research Assistant

Sep 2023 – May 2025 · 1 yr 8 mos · Amherst, Massachusetts, United States

Applied Twitter text-specific pre-trained RoBERTa and Sentence Transformers to analyze 7 years of poll data, using hierarchical and OSLOM graph-based clustering for effective topic segmentation.
Created sentiment analysis to cluster public opinions from Twitter polls, enhancing topic identification.

Natural Language Processing (NLP)Large Language Models (LLM)Data AnalysisCluster Analysis

Cherish

Artificial Intelligence Intern

Jun 2024 – Dec 2024 · 6 mos · Boston, Massachusetts, United States · On-site

Published and presented an approach for LLM time-series forecasting method that encodes temporal structure via overlapping patch segments and optional nearest-neighbor augmentation for richer context.
Achieved ~19% lower MAE on Weather and Traffic datasets vs. the S²IP-LLM baseline across different horizons.

Aspect ratio

Data Scientist

Jan 2023 – Jul 2023 · 6 mos · Pune District, Maharashtra, India

Compiled a master database with publicly available 230 social factors and identified high-risk US counties for readmission rates using K-means clustering and correlation algorithms.
Provided detailed reports of Merck products at different periods to the marketing team resulting in a 15% increase in vaccine sales using SQL to retrieve data, PPT and Excel for better visualization, and VBA to automate repetitive reports.

K-means clusteringMachine LearningData AnalysisSQL

Margrakshak technologies llp

Founder

Jan 2022 – Jul 2022 · 6 mos · India

Founded the company as a designated partner under Start-up India with Certificate No. DIPP104093
Drafted the business model and pitched into getting incubated with Maker Village and being funded 2.5 Million by the Government of India for product development and company’s expansion

Start-up VenturesInternet of Things (IoT)LeadershipOpen Innovation

Zomato

Machine Learning Engineer

Sep 2021 – Dec 2022 · 1 yr 3 mos

Rider Impersonification
Built a TensorFlow model to ensure correspondence between the rider registered and the rider delivering the food
Performed quantitative and qualitative analysis to detect and recognize faces and produce embeddings in under 500ms
Currently, this model impacts 200K riders daily and has brought down the impersonification fraud from 15% to 1%
Device fraud detection
Contributed to solving device fraud using Sentence Transformer and Siamese network to encode input features into vectors and then using a similarity search algorithm to link 2 devices if they are similar
Post linking, an anomaly detection, Isolation Forest model runs, which flags fraud users with a False Positive of 4%
Impacts user experience of 6 Million people daily and receives 2000 Requests Per Second at peak hours
Speech Recognition
Designed a complete pipeline to transcribe Zomato call recordings to enhance user experience.
Researched and implemented a fine-tuned Hugging Face model to classify audio language, then an ASR model with a WER (Word Error Rate) of 23% to transcribe audio files, and finally translate it to english text with a BLEU score of 68.

TensorFlowSiamese networkIsolation ForestSpeech RecognitionMachine LearningArtificial Intelligence (AI)

Mastercard

Software Development Intern

May 2021 – Jul 2021 · 2 mos

Created an Auto Test Case generation module for APIs that is capable of creating a Zero-Code compatible Payload using JSON and hence saves manual labor work
Implemented it by using Spring Boot to code and Postman to hit APIs and check for the payloads

Ivlabs, vnit

2 roles

Vice Chairman

Jun 2020 – Apr 2022 · 1 yr 10 mos

Organized several events spanning talks, discussions, guest lectures, and teaching sessions for over 500 juniors
Drafted and compiled an annual report detailing each of the labs project thoroughly to secure funding from the college

LeadershipManagementOrganization SkillsCommunication