Machine learning · MLOps · Software

VIT Chennai / Class of 2028

Ujjwal Shukla.

From data and models
to useful applications.

I’m a Computer Science student building machine learning pipelines, source-grounded AI tools, and web applications—with an interest in how they work beyond the notebook.

Computer Science student, graduating in 2028Based in Chennai, India

01 / Selected work

A closer look at the work.

Three projects across the ML lifecycle,
retrieval, and geospatial data.

01 Machine learning & MLOps

Student Performance
Prediction

Taking a regression model from experimentation to a repeatable application.

I built a modular Python pipeline for ingestion, transformation, training, evaluation, and inference. Model comparison and GridSearchCV tuning led to AdaBoost Regressor as the final model.

A Flask application serves predictions, supported by logging, model serialization, and unit tests. MLflow tracks experiments and models; DVC versions the data and model artifacts.

My contribution
ML pipeline, inference application, experiment tracking, and deployment workflows.
Deliverable
A structured ML application with a CodePipeline → Elastic Beanstalk workflow and a separate Docker → Amazon ECR path.

Python · Scikit-learn · Flask · MLflow · DVC · AWS

From training to inferenceWorkflow diagram
  1. 01
    Ingest & transformModular data preparation
  2. 02
    Compare & tuneGridSearchCV · AdaBoost selected
  3. 03
    Evaluate & serializeReusable model artifacts
  4. 04
    Serve predictionsFlask inference application
MLflow experiment trackingDVC data & model versioning

Separate deployment paths

GitHub CodePipeline Elastic Beanstalk

Docker image Amazon ECR

Pipeline and deployment workflows documented in my resume. The container path is separate from Elastic Beanstalk.

02 Retrieval-augmented generation

Fact-Checking
Assistant

Bringing published fact-checks into the answer.

I built a knowledge base from published fact-check articles, including AltNews and BOOM Live, and implemented the retrieval pipeline from document processing to reranking.

Articles are chunked and embedded for vector retrieval. Relevant passages are reranked before being passed to an LLM, grounding the response in the retrieved source material.

My contribution
Article collection, document processing, embeddings, retrieval, reranking, and response generation.
Design decision
Retrieve published evidence before generating an answer. Coverage and source quality still limit the result.

Python · Web scraping · Embeddings · Vector retrieval · RAG

Following the evidenceWorkflow diagram
AltNewsPublished fact-checksBOOM LivePublished fact-checks
Process · chunk · embedBuild the searchable knowledge base
User query
Retrieve & rerankSelect relevant passages
A source-grounded responseLLM + retrieved context
Evidence retrieval supports an answer; it does not guarantee that every claim can be verified.

03 Geospatial data & machine learning

Landroid

A clearer picture of land health, from multiple data sources.

Built with a four-person team, Landroid combines soil, terrain, and geospatial APIs with Sentinel-2 NDVI imagery to assess land health.

I owned the scoring pipeline that brought these inputs into composite land health scores, and contributed to the data visualization and map overlays that made the results easier to interpret.

My contribution
Land health scoring pipeline, visualization, and map overlay components.
Result
2nd place at a national hackathon among 100+ teams.

Python · Sentinel-2 NDVI · GIS · REST APIs

Reading the landWorkflow diagram
ASoil & terrainExternal APIs
BGeospatial dataGIS inputs
CSentinel-2NDVI imagery
My primary ownershipLand health scoringCombine inputs into composite scores
Visualization & map overlaysPresent land health and validation results
Built by a team of fourNational hackathon · 2nd place
Conceptual data flow of the project. This diagram represents the workflow, not a satellite image or product screenshot.

More things I’ve worked on.

GitHub profile

LionScott

A gym management platform covering member management, billing, payment integration, and an admin dashboard. My work spanned the interface and backend.

React · Node.js · Express
Web application

Algerian Forest
Fire Predictor

Exploratory analysis and feature engineering on meteorological data, followed by model comparison and Pickle serialization for a prediction application.

Python · Scikit-learn · Pandas
Machine learning

Hcare MedVigilant

A Streamlit medical cabinet prototype with active-ingredient conflict checks, OCR-based medicine identification, and image-based disease prediction.

Python · Streamlit · Hugging Face · OCRView source
AI application

Wine Dataset EDA

An exploratory analysis of the Vinho Verde wine dataset, documenting methodology, visualizations, and statistical findings.

Python · Pandas · Seaborn · Matplotlib
Data analysis

02 / A little about me

Curiosity, followed
through in code.

I started with C and C++, then found my way to Python and machine learning. What keeps me interested is the whole path: understanding the data, choosing a model, and making it usable.

That interest has taken me from exploratory analysis to tracked ML experiments, deployment workflows, and retrieval-based AI. I also enjoy the software side—building interfaces and the systems behind them.

At VIT Chennai, I’m continuing to develop that practice through coursework, personal projects, and collaborative hackathons.

More in my resume PDF

Education / 2024–2028

VIT Chennai

B.Tech in Computer Science & Engineering

8.98 / 10 CGPA

Leadership / 2024–present

Microsoft Innovation Club

Member, VIT Chennai

Helped organize and run 5+ large-scale tech hackathons, contributing to logistics, participant coordination, and event operations.

Tools in context

What I work with.

Models & data
Python, Scikit-learn, Pandas, NumPy, model evaluation, and GridSearchCV.In Student Performance Prediction
Retrieval & language
Document processing, embeddings, vector retrieval, reranking, and RAG.In the Fact-Checking Assistant
Applications & deployment
Flask, React, Node.js, Docker, MLflow, DVC, CI/CD, AWS CodePipeline, Elastic Beanstalk, and ECR.Across ML and web projects

03 / Get in touch

Have something
in mind? Let’s talk.

I’d be glad to hear about internships, collaborations, or a project you’re working on.