Utkarsh Singh

AI Researcher · Developer

Second-year Computer Science Engineering student at MAIT (GGSIPU), specialising in AI. I build practical AI tools across agentic workflows, backend infrastructure, and applied machine learning — driven by the craft of turning ambitious ideas into working systems.

Published researcher at ICML 2026 on Adaptive Termination in Agentic Multimodal RAG — designing CMEGS, a claim-level grounding metric, and MIR, a principled stopping criterion that tells a retrieval agent when it actually knows enough.

Get in Touch View Research

About

I'm a Second-year B.Tech student in Computer Science Engineering (AI specialisation) at Maharaja Agrasen Institute of Technology, GGSIPU — currently maintaining a 9.3 CGPA.

My work spans agentic AI systems, multimodal retrieval, and full-stack ML applications. I'm particularly interested in how AI agents decide what they know — and when to stop looking.

Outside coursework, I build systems that are meant to work in the real world, not just notebooks. That's led to projects in automotive AI, health tech, and enterprise code intelligence.

Education

B.Tech in Computer Science Engineering

Specialisation in Artificial Intelligence

Maharaja Agrasen Institute of Technology, GGSIPU

New Delhi · Aug 2025 – May 2029 · 9.3 CGPA

Class XII

PCM with Computer Science

Kamal Public Sr. Sec. School, CBSE

New Delhi · 2025 · 92.4%

Research

ICML 2026

Adaptive Termination in Agentic Multimodal RAG via Cross-Model Evidence Grounding

International Conference on Machine Learning — ~26% global acceptance rate — Published as a first-year undergraduate

Most retrieval-augmented systems retrieve greedily or stop arbitrarily — they don't know when they know enough. This paper introduces two contributions: CMEGS (Cross-Model Evidence Grounding Score), a claim-level grounding metric that measures whether retrieved evidence actually supports each assertion; and MIR (Minimum Information Requirement), a principled stopping criterion that gives a multimodal agentic RAG pipeline a mathematically grounded sense of sufficiency.

Read Paper →

drafting more...

Projects

Fullstack · ML

ATLASS

Adaptive Taste Learning And Suggestion System — a full-stack recommendation engine that uses collaborative filtering (NMF/SVD on MovieLens) to suggest movies and series tailored to individual taste, with a modular frontend architecture separating data pipelines, model integration, and UI.

ReactPythonNMF/SVDscikit-learn

AI Shopping Agent · E-commerce

SCOUT

An AI-powered cross-platform shopping assistant for India that searches Amazon, Flipkart, Blinkit, Zepto, Instamart, Myntra and Ajio simultaneously, ranks results through a six-factor scoring engine, and streams AI-generated recommendations in real time via SSE.

PythonFastAPIClaude APISSERedis

ML · Automotive · India

TRACTION

Telemetry-based Real-time Automotive Intelligence Platform — an ML-powered vehicle co-pilot for Indian mid-range cars, translating raw sensor data across 47 fields into plain-language advisories. Built on a synthetic dataset of 10,800 readings across 180 Indian driving sessions, accounting for Delhi/NCR traffic patterns and fuel adulteration scenarios.

PythonMLStreamlitPandas

Fullstack · Health · GenAI

CareFlow

A full-stack personal health dashboard covering nutrition, hydration, medication reminders, BMI tracking, and real-time geospatial mapping of nearby medical facilities via Leaflet. Gemini AI powers an intelligent health assistant for personalised queries. Built with Framer Motion and GSAP micro-interactions.

Express.jsGemini AILeafletGSAPFramer Motion

Currently Working On →

Skills

AI & Machine Learning

Generative AIAgentic AIMulti-Agent Systems RAGMultimodal RAGLLM Applications Deep LearningPyTorchTensorFlow TransformersHugging Facescikit-learn

Languages

PythonTypeScriptJavaScript SQLC++C

Backend & APIs

FastAPIFlaskNode.js Express.jsREST APIs

Frontend

ReactNext.jsTailwind CSS

Databases

PostgreSQLMySQLMongoDB RedisFirebaseSQLite

MLOps & Cloud

DockerGitGitHub Actions (CI/CD) AWS (S3, EC2, Lambda)Google Cloud Platform VercelRenderRailway

Libraries & Tools

NumPyPandasLangChain MatplotlibOpenCVPostman Jupyter

Achievements

ICML 2026 · Published Researcher

Published at International Conference on Machine Learning

Co-authored a paper accepted at ICML 2026 — one of the most selective ML venues globally at ~26% acceptance rate. Published as a first-year undergraduate.

November 2025 · National Honour

Award of Excellence

Honored with the Academic Excellence Award, presented by Shri Suresh Jain(National Organization Secretary, Bharat Vikas Parishad) and Shri Vijay Kumar(Joint Commissioner of Police).

May 2026 · National Hackathon

Recognised as Top Performing Candidate — ET Gen AI Hackathon

Co-built an Autonomous Multi-Agent AI Code Intelligence & Remediation Platform. Ranked Top Performing candidate out of 3,000+ participants nationwide at one of India's largest GenAI hackathons.

April 2026 · Open Source

Open Source Contributor — GSSoC & SSoC

Selected as a contributor for GirlScript Summer of Code and Social Summer of Code, contributing to community-driven software projects.

locked in for more

Certifications

AWS Fundamentals of Machine Learning and Artificial Intelligence Jan 2026

IBM SkillsBuild Getting Started with Generative AI Jan 2026

Economic Times Top Performing Candidate — ET GenAI Hackathon May 2026

More Certificates