Jestin
Thomas

AI/ML Engineer & Data Scientist

MSc Data Science · IEEE Published Researcher · Government of India Design Patent Co-holder

0%

Peak Accuracy

0K+

Data Points

0

Projects

0

Internships

Python
CNN
LLMs
SQL
RAG
AWS
Power BI
Docker
TensorFlow
Streamlit
Jestin Thomas

Building AI that
works in the real world

MSc Data Science graduate from CHRIST University (2026) with a BSc in Statistics and hands-on AI/ML engineering experience across internships and independent production-grade projects. IEEE-published researcher and Government of India Design Patent co-holder.

Experienced in end-to-end machine learning and deep learning — from data preprocessing, feature engineering, and model training to deployment, monitoring, and retraining. I build Generative AI applications, LLM-powered systems, RAG pipelines, and agentic AI workflows using LangChain, LlamaIndex, and model APIs.

Strong statistical foundation with applied experience in hypothesis testing, A/B testing, predictive analytics, and data-driven decision making. I follow MLOps best practices with CI/CD, experiment tracking, and model versioning.

PythonTensorFlowPyTorchLangChain RAGLLMOpsPower BISQL AWSDockerStreamlitGitHub Actions Scikit-learnApache Spark

Machine Learning & AI

CNN, LSTM, GRU, Transformers, XGBoost, Random Forest, Transfer Learning, A/B Testing, Time Series

Generative AI & LLMOps

LLMs, RAG, AI Agents, LangChain, GANs, VAE, Diffusion Models, Prompt Engineering, FAISS

Data Science & Analytics

EDA, Hypothesis Testing, Predictive Analytics, Statistical Modelling, Power BI, Tableau, Plotly

Cloud & Infrastructure

AWS, Google Cloud, Docker, Flask, FastAPI, GitHub Actions CI/CD, MLOps, Streamlit Cloud

Where I've worked

TechZia

Jan 2026 – Jun 2026

AI/ML Engineering

AI/ML Engineer Intern

Autonomous Vehicle Cybersecurity Agent

  • Architected CNN + LSTM real-time anomaly detection on AV telemetry (IMU, CAN bus, GPS) in ROS 2 Humble across 3 sensor streams with sub-100ms inference latency
  • Built a DQN-based multi-agent reinforcement learning system with 3 specialised RL agents under a hierarchical global decision agent for sensor security, vehicle behaviour, and network cybersecurity
  • Preprocessed the comma.ai AV dataset via vectorised NumPy pipelines and researched CNN-Transformer hybrid and GNN-based multi-sensor fusion architectures for network intrusion detection
PythonTensorFlowROS 2CNN+LSTMDQNMARLGNN

Seeroo IT Solutions

Apr 2025 – May 2025

Data Science

Data Science Intern

100-Hour Internship · Kalamaserry, Kochi

  • Optimised NLP query-processing pipelines, achieving a 62% reduction in average query latency and enabling the system to reliably handle 500+ daily queries
  • Built and deployed an intelligent chatbot for data-related support queries, reducing analyst response load using Python, Pandas, and Scikit-learn
  • Performed end-to-end EDA and feature pipeline standardisation across complex multi-source datasets to improve downstream model accuracy
PythonPandasScikit-learnNLPEDAChatbot

Things I've built

01 GitHub

Automated Species Identification

CNN transfer learning on 10,000+ images across 90 species. Boosted accuracy from 35% to 91.76% via hyperparameter tuning. Production REST API at 100+ req/sec with under 200ms latency.

91.76%Accuracy
90Species
<200msLatency
ResNet50VGG16TensorFlowFlaskREST API
02 GitHub

Solar Flare Classification

DenseNet121 + GRU + Attention model trained on 8,000+ satellite sequences, achieving 75% accuracy with 58% M-class recall. Deployed in React via TensorFlow.js with no backend. Research published on IEEE Xplore.

75%Accuracy
8K+Sequences
IEEEPublished
DenseNet121GRUAttentionTF.jsReact
03 GitHub

Retail Profitability Intelligence

End-to-end EDA pipeline with Spark and SQL aggregations. Engineered Profit Margin and Risk Score KPIs for discount impact analysis. Delivered interactive Power BI and Tableau dashboards for pricing decisions.

5+KPIs Built
SparkProcessing
Power BIDashboards
PythonSQLApache SparkPower BITableau

Zomato Market Intelligence Platform

Restaurant market intelligence platform analysing 12,000+ Bengaluru restaurants across 90+ locations. Covers EDA, sentiment analysis, Random Forest rating prediction, K-Means clustering, and a Market Expansion Advisor deployed on Streamlit Cloud.

12K+Restaurants
90+Locations
R²=0.41Rating Model
PythonPandasScikit-LearnRandom ForestK-MeansStreamlitPlotly

Model Regression Detection LLMOps

LLMOps platform that automates LLM evaluation, monitoring, and regression detection. Benchmarks Groq and Gemini using a golden dataset, compares prompt versions, and runs full CI/CD via GitHub Actions on every push.

Groq+ Gemini
CI/CDAutomated
AutoRegression Det.
PythonLLMOpsGroq APIGemini APIStreamlitGitHub Actions
06 GitHub

Multi-Document RAG Research Assistant

AI-powered Q&A system for multiple research PDFs using FAISS vector indexing and Sentence Transformer embeddings. LangChain RAG pipeline with RAGAS evaluation and Weights & Biases experiment tracking to minimise hallucinations.

FAISSVector Search
RAGASEvaluation
W&BTracking
LangChainLlamaIndexFAISSRAGSentence TransformersRAGASStreamlit

Academic background

CHRIST University
2024 – 2026 · Completed

Master of Science in Data Science

CHRIST (Deemed to be University)

Bengaluru, India

Machine LearningDeep LearningNLPStatistical Modelling
MG University
2021 – 2024 · Completed

Bachelor of Science in Statistics

Sree Sankara College, Kalady

MG University, Kerala

StatisticsProbabilityData AnalysisR Programming

Publications & patents

IEEE Xplore Feb 2026 · ICVADV-2026

Synthetic Data Augmentation for Robust Solar Flare Classification

A Comparative Analysis of Conditional DCGAN, VAE, and Diffusion Models

Benchmarked three generative deep learning architectures for improving classifier robustness on imbalanced astronomical datasets. Demonstrated measurable improvement in minority class recall through synthetic augmentation.

DCGANVAEDiffusion ModelsClass Imbalance
Read on IEEE Xplore
Govt. of India Jan 2025 · No. 443142-001

Mobile Cooler — Design Patent

Class 23-04 · Co-inventor

Co-inventor of a Government of India registered Design Patent for a Mobile Cooler device. Demonstrates innovation that extends beyond software into physical product design and intellectual property.

Industrial DesignPatentClass 23-04
View Patent Document

Open to
new opportunities

Looking for full-time roles in AI/ML Engineering, Data Science, and Data Analytics. Based in Bengaluru and open to hybrid or remote work.

Download Resume

Send a Message

Messages go directly to my Gmail.