Aryan Mhalsank · Computer vision engineer

I teach machines
how to see.

I build visual AI systems and explore how machines can learn richer representations from less data — with research that keeps people in view.

Illustrated portrait of Aryan wearing a black Punjabi Sikh turban and holding a computer-vision camera module object / 0.98depth map · liveCAM_01   TRACKINGCurious by nature.
Research-minded by choice.
Scroll to explore

A practice at the intersection of

perception, learning & human understanding.

Based in Mumbai, IndiaOpen to research · collaboration · graduate study
01 / Selected work

Building systems that see.

A selection of applied computer-vision and AI projects — from pixels on the edge to multimodal research tools.

Explore by focus
06 studies
Computer-vision PPE safety scene showing a worker in a high-visibility vest and hard hat on a factory floor02 / VISUAL STUDY
Video analytics · Workplace safety02

PPE Detection

Safety, understood in real time

A multi-camera monitoring system for detecting helmet, vest and glove compliance across factory feeds.

APPROACH

Custom YOLOv8 (5 classes) → DeepSORT tracking → alert webhook

YOLOv8OpenCVDeepSORTFlask

↗91% precision · 88% recall · 4 simultaneous cameras

Retinal fundus image and magnified research crops with subtle segmentation contours03 / VISUAL STUDY
Medical imaging · Self-supervised learning03

RetinaSense

Learning from unlabeled images

A self-supervised retinal disease classification study, pretrained on unlabeled fundus imagery before supervised evaluation.

APPROACH

SimCLR backbone → ViT-S → linear probe → 5-class classifier

PyTorchSimCLRViTAlbumentations

↗AUC 0.91 · 3× less labeled data needed

Multimodal AI · Research understanding04

VigyaanSetu

Research papers into something you can explore

An open-source platform for asking questions across research papers and transforming the material into narrated video lectures.

APPROACH

PDF → OCR → BGE embeddings → RAG → LLM → TTS → video

LangChainFAISSGPT-4oWhisperMoviePy

↗2 publications · IEEE in-proceeding · open source

Agentic systems · Cloud optimization05

OptiSave.AI

Finding room to do more with less

A multi-agent platform that reasons over cloud usage telemetry to recommend cost and storage optimizations.

APPROACH

Lyzr analyzer / planner / executor agents → cloud APIs

LyzrPythonAWS SDKPandas

↗Winner — Buildathon 2.0 by 100x Engineers

Medical imaging · Cross-modal learning06

Thermal × Mammographic Fusion

Two imaging modalities, one perspective

A self-supervised bi-pipeline exploring fusion of thermal and mammographic images for breast cancer detection.

APPROACH

Dual-encoder SimSiam → cross-modal fusion → classifier head

PyTorchMONAIOpenCV

↗Published — Springer Nature

02 / Research direction

Questions worth staying with.

My long-term goal is graduate research in self-supervised and multimodal representation learning — connecting visual perception with a deeper understanding of the physical world.

THE QUESTION

How can machines learn useful visual representations with less annotation — and carry that understanding across modalities, settings and people?

Looking toward PhD / MS-by-Research opportunities
AREAS I’M DRAWN TO
01Self-supervised and contrastive representation learning
02Vision–language grounding, multimodal models and VQA
033D perception, depth and scene understanding
04Medical imaging, image restoration and fusion
05Assistive vision and wearable edge inference
06AI systems for scientific discovery
Fundus image and magnified areas illustrating medical image analysis
MEDICAL IMAGINGSelf-supervised representation learning
A hazy city view compared with its contrast-restored version
IMAGE RESTORATIONDehazing · Indian Statistical Institute
SELECTED PUBLICATIONSResearch that travels beyond the lab
01

VigyaanSetu: AI-Powered Platform for Research Paper Understanding & Multimodal Knowledge Transformation

Research understanding
In proceedings
02

A Framework for AI-Driven Research Automation: The VigyaanSetu Approach

Research automation
IEEE · in proceedings
03

Thermal & Mammographic Image Fusion for Breast Cancer Detection — Self-Supervised Bi-Pipeline

Medical imaging
Springer Nature
04

Detecting Heart Disease Using Machine Learning

Applied machine learning
IEEE
03 / Experience

Learning by making.

A path through software, open source and research — each step bringing computation closer to the visual world.

01
Jan — Oct 2026

Nirikshan AI

AI Intern

Built computer-vision pipelines with YOLO fine-tuning, OCR extraction and a real-time text-to-speech overlay.

YOLOOCRTTS
02
Jun 2024 — May 2025

Indian Statistical Institute

Research Intern

Explored self-supervised image enhancement for low-light and medical imagery, alongside a single-image dehazing research project.

Self-supervisionImage enhancementDehazing
03
May — Aug 2024

Social Summer of Code · S3

Open-source contributor

Contributed to open-source computer-vision and machine-learning projects.

Open sourcePythonGit
04
Jun — Aug 2022

Panel Technologies

Software Development Intern

Built invoice-digitization software, an early introduction to document processing and applied machine learning.

PythonAutomation
A FEW MILESTONES
01

Winner — Buildathon 2.0, 100x Engineers

02

Top 5 — LLM Hackathon at IIT Bombay

03

Agri-Tech Hackathon — KJCSE, national level

04

Runner-up — state-level project competition

WHERE I STARTED

B.E. · Artificial Intelligence & Data Science

MCT’s Rajiv Gandhi Institute · 2023–2026

Diploma · Computer Engineering

Vidyalankar Polytechnic · 2020–2023

04 / Toolkit

From pixels to pipelines.

Tools I reach for across perception, representation learning and full-system delivery.

Vision

OpenCVYOLOObject detectionSegmentationOCRVideo analytics

Learning

PyTorchTensorFlowSelf-supervisionSimCLRVision transformers

Multimodal

LLMsRAGVision–languageWhisperTTS

Building

PythonDjango / FlaskReactLinuxEdge inference
05 / Start a conversation

Good research starts
with a good question.

I’m open to research opportunities, thoughtful collaborations and conversations about computer vision.

aryanmhalsank27@gmail.com +91 92213 31197