Harshit Shah

Intern

Mumbai, Maharashtra, India1 yr 6 mos experience

Most Likely To Switch

Key Highlights

Guinness World Record holder for solar lamps project.
Active contributor to Atal Tinkering Lab promoting STEM skills.
Developed advanced ML models for security applications.

Stackforce AI infers this person is a Fintech and AI specialist with a focus on data science and machine learning.

Contact

Skills

Core Skills

Machine LearningQuantitative AnalyticsStatistical Data AnalysisDeep Learning

Other Skills

C++Core JavaCryptocurrencyData ScienceData StructuresDatabase Management System (DBMS)DatabasesImage ProcessingJavaJulia (Programming Language)MicroservicesMySQLObject DetectionOpenCVOrganization Skills

About

• Recognized by Guinness World Records in 2018 at IIT-Bombay for leading a successful team effort in setting the record for "Most Solar Lamps Lit Simultaneously for 5 Minutes." • Contributed actively to the Atal Tinkering Lab, an initiative supported by ISRO, aimed at encouraging innovation and hands-on learning among students. Helped promote skills like design thinking, computational logic, and physical computing, supporting the lab's goal of building a strong foundation in STEM. - Always up for a convo on Tech & Finance

Experience

Undisclosed

Intern

Mar 2025 – Present · 1 yr

Quantifying uncertainty in complex, dynamic systems.

Devine group

Quant Engineering & Analytics Intern

Feb 2025 – Apr 2025 · 2 mos · London Area, United Kingdom · Remote

We're a quantitative finance research & development firm specializing in statistical arbitrage, quantitative research, and DeFi (Decentralized Finance).
Built end-to-end pipelines and set up multi-venue crypto trading for CEXs to support statistical arbitrage strategies.
Worked on performance & risk analytics to evaluate trading strategies.

Numerai

Quant Data Scientist - Contributor

Oct 2024 – Present · 1 yr 5 mos · San Francisco, California, United States · Remote

We're a hedge fund, built by a network of data scientists, managing an institutional grade long/short global equity strategy for our investors.
Building ML models on the highest quality market data, alongside a team of the best data scientists from around the world.
Checkout:
https://blog.numer.ai/jpmorgan-secures-500m-capacity/

Machine LearningQuantitative AnalyticsData Science

Scale ai

Data Scientist - Contributor

Sep 2024 – Feb 2025 · 5 mos · San Francisco, California, United States · Remote

A subsidiary created and managed by Scale AI (valued at approx. $29 billion USD as of June 2025 after Meta's investment of $14.3 billion USD)
Helped SOTA models improve their math reasoning capabilities.

Kapidhwaj ai

Machine Learning Engineering Intern

Sep 2024 – Nov 2024 · 2 mos · India · Remote

Developed next-gen ML models & end-to-end pipelines, integrating ML with CCTV systems to enhance security and operational efficiency.
Built and optimised custom + open-source ML models for the following services:
i.) Face recognition: Accurately detected and identified faces in live video streams.
ii.) Intrusion detection: Enhanced security by detecting unauthorised access with high accuracy.
iii.) Motion detection: Improved accuracy from ~75% to ~90% by migrating from frame differencing to background subtraction and using advanced techniques to detect human and object motion effectively.
More about the motion detection model here:
https://docs.google.com/document/d/1FHxx3ke-vk8lE3jKqJ9HlHtZuYm8UGhAAjPBsqO6wOs/edit?usp=sharing
Lastly,
Implemented features like timestamped motion alerts and automated video recording during detections.
Integrated cloud storage workflows with Google Cloud Storage (GCP) to store processed images and videos securely.

University of oxford

Capstone Project - AI & Deep Learning

Mar 2024 – May 2024 · 2 mos · Oxford, England, United Kingdom · Remote

Advisor: Prof. Pramit Saha
Developed advanced fine-grained image classifiers using EfficientNet and ConvNeXt, improving accuracy by 15% and reducing inference time by 10%.
Optimised image processing pipelines with EfficientNet and ConvNeXt, leading to a 10% increase in model efficiency.
Achieved significant performance improvements with EfficientNet and ConvNeXt architectures, boosting classification accuracy by 22% and reducing training time by ~10%.