AI Engineer & Researcher · Mumbai, India

SAMPURNA
ROY

Specializing in  

AI/ML Engineer with published research in IEEE, Taylor & Francis, and ETASR. Building award-winning AI frameworks that push the boundary between research and production. Currently working as an AI/ML Developer (Design Engineer) at Selec Controls.

View My Work GitHub
3
Publications
98.2%
AI Accuracy
10
GitHub Repos
8.54
CGPA
Sampurna Roy
3
Publications
98.2%
AI Accuracy
10
GitHub Repos
8.54
CGPA
Scroll
// 01

ABOUT ME

I'm Sampurna Roy — an AI/ML Engineer and researcher who lives at the intersection of deep learning, systems engineering, and applied AI, currently working as an AI/ML Developer at Selec Controls.

My research has been published in IEEE, Taylor & Francis, and ETASR, with a Best Paper Award at IEEE ICETET-SIP-25 for my Alzheimer's prediction framework achieving 98.2% accuracy using RL + GNNs.

I combine strong programming fundamentals with applied AI/ML expertise to design, build, and ship reliable, scalable products — from computer vision pipelines for industrial robots to LLM-powered enterprise automation portals.

AI / ML Deep Learning NLP & LLMs Computer Vision Graph Neural Networks Reinforcement Learning Python Research
Sampurna Roy
Amity University MumbaiB.Tech CSE (AIML) · Aug 2022 – June 2026 · CGPA 8.54
Lodha World School (ICSE)12th Science 88% · 10th ICSE 90.4%
// 02

WORK EXPERIENCE

AI/ML Developer (Design Engineer)

Selec Controls Pvt. Ltd.
Jul 2026 – Present

AI-Powered Chatbot Suite

  • SELTALKS (Multimodal RAG Assistant): Built with Flask, Gemini, LangChain, and FAISS to diagnose issues across Selec's product catalog from text, error codes, and photos of faulty hardware; Gemini Vision OCRs wiring diagrams, pinouts, and tables from PDFs/Excel/Word docs into the vector store.
  • LLM-as-a-Judge & Enterprise Governance: Added an LLM-as-a-Judge layer scoring every response (0–100%) against retrieved context, ticket escalation with email alerts, an admin "Teach AI" loop that injects verified corrections into the live index without a restart, plus multi-key rotation, role-based auth, and prompt-injection guards.
  • GTI Debug & Service Support Copilot: Streamlit diagnostic assistant for Grid Tie Inverter operators and field engineers; exact matching + scikit-learn TF-IDF / cosine search over debug SOPs and field logs returns checklists, tools, and suspect components; technician-verified fixes are learned and promoted.
  • Production QC Assistant: Text-to-SQL pipeline (LangChain + Groq) turning natural-language questions into SQL over SQLite to avoid hallucinated figures on defect/ATE data; auto-syncs the production Excel workbook, with in-app defect reporting and role-based access (bcrypt).
Python Gemini Groq LangChain RAG FAISS Text-to-SQL Flask Streamlit SQLite

AI-Based PCB Wiring & Solder Inspection System

  • Computer Vision Fault Detection: Detects wiring faults and solder defects from PCB images — OpenCV (HSV masking, contour detection, spatial analysis) verifies red/black wire origins and routing, and a YOLOv8 module detects and counts solder balls/valleys across multi-angle images.
  • Automated Inspection Pipeline: Automated Pass/Fail pipeline over multiple detection modules, with a Tkinter GUI for visual fault indication and bounding boxes; tested and validated on 1,000 units.
Python OpenCV YOLOv8 Computer Vision Tkinter PyQt5/6 Git

Python Developer Intern

Roboserv4i
Feb 2026 – June 2026
  • Developed AI-powered automation solutions for the DuctBot industrial robotic inspection system as a Software, Python & AI Developer Intern.
  • Built and integrated a real-time Computer Vision pipeline using YOLO and OpenCV for defect detection, image processing, and visualization of AI predictions on live video feeds.
  • Designed a Kivy-based Python GUI for real-time robot control, inspection monitoring, and multi-camera streaming.
  • Designed and managed databases using SQLite and Amazon RDS to persist inspection data, AI outputs, metadata, and system logs for end-to-end traceability.
  • Optimized AI inference and application performance on Raspberry Pi through multi-threaded processing and implemented secure, encrypted data handling mechanisms.
Python YOLO OpenCV Kivy Raspberry Pi SQLite Amazon RDS

Software Engineer Intern (Technology Intern)

FoundIt
June 2025 – Aug 2025
  • Contributed to a full-stack e-commerce application using Java (Spring Boot), HTML, CSS, JavaScript, and MySQL, implementing backend REST APIs, user authentication, shopping cart, and product management modules.
  • Developed and integrated an AI-powered chatbot using Google Gemini that delivered personalized product recommendations based on user events and occasions.
  • Analyzed customers' purchase history to generate shopping insights and suggest products tailored to individual buying patterns.
  • Built an interactive analytics dashboard to visualize user spending across product categories, helping users track expenses and make informed purchasing decisions.
Java Spring Boot Google Gemini JavaScript MySQL REST APIs
// 03

PROJECTS & CODE

01
Best Paper Award · IEEE ICETET-SIP-25

Alzheimer's Prediction via RL & Graph Neural Networks

Hybrid AI framework (RL + Spatio-Temporal GNNs + VAEs + SimCLR) achieving 98.2% accuracy on the ADNI dataset for early-stage Alzheimer's detection. Published in IEEE.

02

Bird Sound Classification with Deep Learning

Audio-based bird species classification on the BirdCLEF dataset using MFCC feature extraction and CNNs. ~75% validation accuracy with audio preprocessing and data augmentation.

03

NLP-Based ATC-Pilot Communication System

Real-time communication system using NLP to automate message handling between Air Traffic Controllers and Pilots, with live data updates for enhanced system responsiveness.

04

ScoutAI — Job Search Agent

An intelligent AI-powered job search agent that automates the process of finding, filtering, and surfacing relevant job listings based on user profile and preferences.

05

Disease Prediction System

ML-powered system that predicts diseases based on patient-reported symptoms, enabling early identification and medical guidance through an intuitive interface.

// GITHUB REPOSITORIES
Sampurna Roy GitHub

Sampurna1524

github.com/Sampurna1524  ·  10 public repositories

ScoutAI
Job search agent — AI-powered automation for job hunting
Python
bird-sound-detector
BirdCLEF CNN-based bird species classifier using MFCC features
Python
Disease-Prediction
Predicts disease based on the symptoms faced by the patient
HTML
Customer-Churn-Prediction
ML model that predicts customer churn for business analytics
Jupyter Notebook
Spam-message-detection
NLP-based spam message classifier using machine learning
Jupyter Notebook
Data-Science-Capstone-SpaceX
IBM Data Science Capstone: SpaceX launch data analysis & prediction
Jupyter Notebook
Quizzy
Python & HTML-based MCQ Quiz web application
PythonHTML
Cart-app
Java & HTML-based e-commerce shopping cart web application
JavaHTML
View All Repos on GitHub
// 04

RESEARCH PUBLICATIONS

01
IEEESeptember 2025 Best Paper Award

"Reinforcement Learning and Spatio-Temporal Graph Neural Networks for Alzheimer's Disease Progression"

Hybrid AI framework integrating RL, ST-GNNs, VAEs, and SimCLR for Alzheimer's progression modeling using longitudinal MRI data. Achieved 98.2% accuracy. Best Paper (Track 6) at IEEE ICETET-SIP-25 International Conference.

Read on IEEE Xplore
02
ETASRFebruary 2026

"NeuroFusion for Robust and Explainable Multimodal Deep Learning in Fine-Grained Staging of Alzheimer's Disease"

Interpretable multimodal DL framework for four-stage Alzheimer's staging integrating structural MRI and clinical biomarkers. Combined CNN/Transformer models with XGBoost, GradCAM and attention mechanisms for interpretability.

Read on ETASR
03
Taylor & FrancisJune 2025

"Neuro-Oncology Reimagined: Tailored Prognosis for Brain Tumors Using Adaptive Machine Learning"

ML-based system for improved prediction and prognosis of brain tumor cases using adaptive machine learning techniques for personalized clinical outcome modeling.

Read on Taylor & Francis
// 05

TECH STACK

Languages
Python95%
Java80%
C / C++75%
SQL80%
Kotlin68%
AI / ML
Deep LearningLLMsNLPComputer VisionReinforcement LearningGraph Neural NetworksGenerative AIPrompt EngineeringFeature EngineeringClaude · Gemini · GPTPerplexity
Frameworks
TensorFlow / Keras90%
PyTorch85%
Spring Boot72%
OpenCV / YOLO85%
Data & Tools
PandasNumPyMatplotlibSeabornTableauLibrosaScikit-learnMySQLSQLiteAmazon RDSGit / GitHubJupyter
IBM
IBM Data Science Professional Certificate
IBM · Coursera · ID: 9NDQ5E4HU188
Comprehensive 12-course professional specialization covering Data Science, Python, SQL, Machine Learning, Data Visualization, and Generative AI.
OCI
Oracle Cloud Infrastructure 2025 Certified AI Foundations Associate
Oracle University
Certified proficiency in Oracle Cloud Infrastructure AI services, Generative AI architecture, Machine Learning pipelines, and enterprise LLM deployment.
GCP
Analyze Images with the Cloud Vision API
Google Cloud · ID: 5099568
Hands-on skill badge covering Google Cloud Vision API, image feature extraction, OCR text detection, label detection, and visual search integration.
BCG
GenAI Job Simulation
Boston Consulting Group (BCG) · ID: AwWdRFMDz3iJsX7Fq
Applied generative AI algorithms, automated financial data extraction via Jupyter, and structured prompt engineering for executive decision support.
TATA
GenAI Powered Data Analytics Job Simulation
Tata Group · ID: gMDq953NgNrLX6yd2
Leveraged GenAI workflows to accelerate exploratory data analytics, automate reporting, and formulate strategic business recommendations.
TATA
Data Visualisation: Empowering Business with Insights
Tata Group · ID: CAFmPuyeSut8AhsKX
Created executive-level business intelligence dashboards and visuals communicating complex operational metrics clearly to C-suite leadership.
DLT
Technology Job Simulation
Deloitte Australia · ID: MMjtCFyrC3gKqYgxJ
Simulated enterprise technology consulting workflows, architecture assessment, code analysis, and solution delivery frameworks.
DLT
Data Analytics Job Simulation
Deloitte Australia · ID: YEhNnE4cFHPZuPPq4
Built analytical pipelines, Tableau business intelligence dashboards, and client telemetry performance models.
SKL
Implementing Deep Learning: Optimized Applications
Skillsoft · ID: 152346743
Advanced techniques for deep neural network optimization, gradient tuning, model compression, and high-performance inference.
SKL
Computer Vision: AI & Computer Vision
Skillsoft · ID: 151398532
Convolutional neural networks, object classification, visual tracking pipelines, and feature map processing.
INF
Infosys Springboard AI & Deep Learning Series
Infosys Springboard
Completed curriculum across Deep Learning, Computer Vision Foundations, AI Types, and Human-Computer Interaction overview.
// 06

ACHIEVEMENTS & AWARDS

Hive 3.0 iHub Pitch Challenge Winner

Recognized for outstanding team pitch and product development.

Best Paper Award – IEEE ICETET-SIP-25

Track 6 winner at International Conference 2025.

3 Research Publications

Published in IEEE, Taylor & Francis, and ETASR.

Star Student of the Month

Recognized for academic excellence in school.

Multiple Merit Certificates

Awarded merit certificates in Computer Science.

IBM Data Science Professional

12-course specialization on Coursera – June 2026.

// 07

GET IN TOUCH

Let's Build Something Extraordinary

Whether it's AI research, product development, or anything in between — I'm always open to new opportunities and collaborations. Reach out and let's talk.