Saumya Bhatt

00 / Journey

2014 → NOW

2014

Started · B.Tech

CHARUSAT University

Education · Gujarat

2017

Software Developer

Milople Tech

First professional role

2018

B.Tech Graduate · Research Asst

CHARUSAT University

GPA 8.2 / 10

2019

Data Science Intern

Excelr Solutions

Anomaly Detection

2019

MS Computer Science begins

University of Texas Arlington

Specialisation · AI

2020

Graduate Research Assistant

University of Texas Arlington

Systems · Team Lead

2021

GRA · ML Research & Thesis

University of Texas Arlington

VLN · Robotics · NLP

2023

MS Thesis Defended & Completed

University of Texas Arlington

GPA 3.4 / 4.0 · Milestone

2023

AI Engineer

Stealth Startup

Confidential · AI Product

2024

Specialist, Data Scientist

Blenheim Chalcot

LLMs · RAG · Fine-Tuning

2024–25

Data Scientist

Zensar Technologies

GenAI · Multi-domain · 6 Projects

NOW

Open for opportunities

India · The world is my oyster

Building something cool

01 / Projects

BUILT FOR REAL
CLIENTS.

14

Professional Projects

Built and delivered at companies — hover any card to reveal the client or organisation

01

GenAI · BFSI

GenAI Document Generation

Automated document generation for banking and financial services clients using Generative AI.

GenAIAzure OpenAIPythonBFSI

Zensar · FISOct 2024

02

RAG · Real Estate

Real Estate RAG Chatbot

GenAI chatbot enabling natural language queries over CSV property datasets using FAISS vector search.

RAGFAISSPythonGemini

Zensar · CBRE2024

03

NLP · Sentiment

Product Review Sentiment Pipeline

End-to-end pipeline categorising enterprise reviews as positive / negative / neutral using ML + GenAI.

NLPScikit-learnGenAI

Zensar TechnologiesEnterprise Client

04

ML · Retail

Customer Segmentation Engine

Segmented users into frequent, occasional, low-engagement and one-time cohorts for targeted campaigns.

ClusteringPythonScikit-learn

Zensar TechnologiesRetail Client

05

RecSys · Healthcare

Hospital Food Recommendation System

Hybrid collaborative filtering engine for hospital food ordering, scaled to production at 80%+ accuracy.

Rec Systems80%+ AccuracyPython

80%+production accuracy

Zensar · Compass HealthHealthcare

06

Architecture · Multi-domain

GenAI Solution Scaling

Designed ML architectures and pipelines for chatbots and document summarisers across multiple domains.

GenAIFastAPIDockerAzure

Zensar TechnologiesMulti-domain

07

LLMs · Fine-Tuning

LLM Fine-Tuning Pipeline

Fine-tuned LLaMA 3 and Mistral using multiple strategies for domain-specific task performance.

LLaMA 3MistralFine-TuningPyTorch

Blenheim ChalcotApr – Jun 2024

08

LLM · Customer Service

Call Quality Evaluation System

Automated LLM-based system to score customer care calls, replacing manual QA at scale.

LLMGPT-3.5Gemini 1.5

Blenheim ChalcotApr – Jun 2024

09

RAG · Evaluation

RAG Pipeline & RAGAS Evaluation

Designed RAG pipeline with systematic chunking and embedding experiments, evaluated end-to-end via RAGAS.

RAGRAGASFAISS

Blenheim ChalcotApr – Jun 2024

10

AI Product · Stealth

Confidential AI System

Worked with an early-stage AI startup to build a product and contributed to several AI-driven ideas and initiatives.

AI ProductStartupGenAI

Stealth Startup2023

11

Research · Robotics · NLP

Vision & Language Navigation

Multimodal AI research training robots to navigate 3D environments via language. Bi-LSTM, RL, backtranslation.

Bi-LSTMImitation RLMatterport3D

Univ. of Texas ArlingtonMS Thesis · 2021–23

12

Artificial Intelligence · Anomaly

Time Series Anomaly Detection

Anomaly detection improving downstream prediction accuracy by ~10%.

AIPythonPandas

+10%prediction improvement

Excelr SolutionsInternship · 2019

13

NLP · Address Validation

Valid Address Detection Model

NLTK-based NLP model validating addresses in unstructured text. 92% accuracy on real-world data.

NLTKNLPPython

92%accuracy

CHARUSAT UniversityResearch · 2018–19

14

RecSys · E-commerce

E-commerce Recommendation System

Full recommendation system for an e-commerce platform — ML model, frontend wireframes and end-to-end documentation.

Rec SystemsPythonJSPHP

Milople TechDec 2017 – Apr 2018

13

Personal Projects

Self-driven experiments — from classical ML to advanced diffusion models

13

// independent research & experimentation

Where curiosity
meets code

Thirteen self-initiated projects spanning the full arc of modern AI — from training classical machine learning models on real-world datasets, to building neural networks for sequence and language tasks, generating photorealistic imagery with advanced diffusion models, and wiring up end-to-end conversational AI chatbots. No client brief, no deadline — just curiosity, a dataset, and the drive to figure it out.

Highlights include: Stable Diffusion Image Generator — generating photorealistic images from prompts by exploring denoising steps and noise schedules; Income Prediction on Census Data — an ensemble of Gradient Boosting, Random Forest and AdaBoost achieving 87% accuracy with full EDA; and Named Entity Recognition Pipeline — extracting and classifying entities from unstructured text at 92% accuracy, visualised end-to-end in Tableau.

13Projects total

5+AI sub-domains

92%Peak accuracy

Classical ML GenAI Neural Networks Diffusion Models NLP Chatbots Computer Vision Data Science

02 / Research

ACADEMIC DEPTH

Master's Thesis · University of Texas at Arlington · 2023

Language Pre-Training and Auxiliary Tasks for Vision and Language Navigation

MS Computer Science — Specialisation in Artificial Intelligence · GPA 3.4 / 4.0

Tackled Embodied AI: training a robot to follow natural language instructions in photorealistic 3D environments. Benchmarked sequence and RL-based navigation strategies, then proposed architectural improvements via richer language pre-training and auxiliary grounding tasks that measurably increased navigation success rates.

Bi-LSTM + AttentionSelf-Supervised ILBacktranslationEnv DropoutPre-trained EmbeddingsAuxiliary TasksMatterport3D

View Thesis ↗

Research domain

Multimodal AI — bridging Natural Language Processing with spatial robot navigation in 3D environments

VLN

Vision-Language Navigation

Prior Research · CHARUSAT 2018–19

In-depth study of collaborative, content-based and hybrid recommendation architectures

92%

Address detection model accuracy

// the person behind the work

BEYOND THE CODE

Saumya Bhatt

I'm someone who genuinely loves to work — not out of obligation, but because building things that matter is where I feel most alive. When I'm not deep in a model or a dataset, you'll find me out in the world, camera in hand.

⚡

Work is a craft, not a job

I bring full focus to whatever I take on. Six years in, the curiosity hasn't faded — if anything it's sharper. I find it hard to do things halfway.

🤝

Good with people

Technical skills get you in the room. People skills get things done. I bridge both — comfortable going deep with engineers and equally at ease translating complexity for stakeholders.

📷

Phone in hand

I don't need a fancy setup — just my phone and the right moment. I'm drawn to candid shots and natural light, finding the frame in everyday scenes that most people walk straight past.

🧭

Exploring places

New cities, winding roads, good food, and the kind of views that make you stop mid-sentence — I'm always looking for the next place to explore. Travel resets me and feeds the curiosity I bring back to work.

"When I'm not working, I'm somewhere figuring out my next adventure"

03 / Articles

WRITING FOR PRACTITIONERS

✦

Thoughtful writing on AI, technology and its role in shaping economies and the environment — accessible reads that go beyond the buzzwords and explore what these shifts actually mean.

01

LinkedIn Pulse · AI & Economics

Impacting Economies: Artificial Intelligence — Beyond the Buzz

AI · PolicyREAD ↗

02

LinkedIn Pulse · Career Guide

The First Step to Becoming a Data Scientist / ML Engineer

Career · MLREAD ↗

03

LinkedIn Pulse · Tech & Society

Technology for the Benefit of the Environment

Tech · ClimateREAD ↗