Open to Software, Data, Product & AI Roles

Software + Data Product + AI

Computer Science engineer who turns problems into working products: defining requirements, analyzing data with SQL and Python, and building full-stack and GenAI systems with FastAPI, React, and PostgreSQL.

View My Work Resume

About

Problem First, Then the Build

I'm a final-year Computer Science student at VIT who likes understanding the user or business problem before writing any code. At Netleaf Software, I translated business requirements for an enterprise HRMS product into user stories, acceptance criteria, and success metrics, then helped build, test, and release what we had specified.

I work across both sides of a product: requirements analysis, documentation, and Agile/Scrum delivery with developers and stakeholders; SQL, Python, Excel, and analytics dashboards for understanding data; Figma for design work; and GenAI for prototypes where it genuinely helps.

How I Work

  1. Understand the problem

    Name the user pain before picking a solution.

    Cortex started from email overload and unclear task priority.

  2. Break down requirements

    Turn a vague ask into stories and criteria a team can build and test.

    At Netleaf: user stories, acceptance criteria, user flows, success metrics.

  3. Analyze the data

    Find where the real friction is, with numbers.

    Measured slow attendance-data responses and cut them from 4.5s to 1.8s.

  4. Prototype and build

    Get a working version in front of real inputs quickly.

    Cortex was tested on real email; Fact Knowledge Layer on four real documents.

  5. Validate and iterate

    Test against reality, then fix what breaks.

    Fact Knowledge Layer: 118 passing tests, and a fix that cut reported contradictions from 107 to 2.

Timeline

Experience & Education

Full Stack Developer Intern

May 2026 — Jul 2026

Netleaf Software · On-site, Gurugram · Enterprise HRMS product

Product

  • Translated business requirements into user stories, acceptance criteria, user flows, functional and QA requirements, and success metrics for employee and attendance workflows.
  • Partnered with developers and stakeholders to clarify requirements, flag delivery risks, and support releases, including post-release support, in an Agile/Scrum team.
  • Analyzed test results and documented test cases and defects in Excel to support product quality improvements.

Engineering

  • Cut attendance-data response time by 60% (4.5s to 1.8s) across 5,000–10,000 records by optimizing SQL queries, search, and pagination in FastAPI and PostgreSQL APIs.
  • Validated features with Postman and load-tested APIs at 50+ concurrent users with JMeter.
  • Implemented JWT authentication and role-based access control across 4 user roles aligned to the organization hierarchy.
FastAPIPostgreSQLSQLJWT / RBACPostmanJMeterExcelDockerGitHub Actions

B.Tech, Computer Science & Engineering

Aug 2023 — Jun 2027

Vellore Institute of Technology · Vellore, Tamil Nadu

  • CGPA 7.8 / 10, no active backlogs.
  • Core coursework in Data Structures & Algorithms, DBMS, Operating Systems, Computer Networks, and Computer Architecture.

Selected Work

Products I've Built

Synced with GitHub

Data reliability

Fact Knowledge Layer

Grounded fact extraction and reconciliation

Aug 2026 — Sep 2026
Grounded facts
680
Reconciled relations
1,446
Documents
4
Passing tests
118

Problem

The same figure shows up differently across reports, with a different scope, period, unit, or scale, so it's hard to tell a real contradiction from a harmless difference.

What I built

A system that extracts facts from PDFs, grounds each one to its exact source evidence, reconciles claims across documents, and names which dimension explains a difference.

Product & technical decisions

  • Let the LLM propose facts, but verify every quote against the source; untraceable facts are quarantined, not silently kept.
  • Made the final verdict deterministic, so the same inputs always give the same, auditable result.
  • Tested on real documents, found distinct claims being merged, and fixed it: reported contradictions dropped from 107 to 2.

Outcome

A working knowledge layer where every stored fact traces back to evidence in its source document, backed by an automated test suite.

PythonSQLFastAPISQLitePyMuPDFGroqGeminiRapidFuzz
Demo video

Productivity analytics

Cortex

AI email intelligence assistant

May 2026 — Present
Coordinated agents
5
Tested on
Real email

Problem

Email overload buries action items, and there's no clear way to see which messages and tasks need attention first.

What I built

An assistant that reads unread Gmail, extracts actionable tasks, prioritizes messages, and writes a daily briefing, plus a Streamlit dashboard tracking task status, email volume, priority distribution, and response trends.

Product & technical decisions

  • Split the workflow into 5 LangGraph agents: supervisor, email, task, priority, and briefing.
  • Grounded task extraction in past email context with ChromaDB-backed RAG and vector search, using Llama 3.
  • Modelled emails and tasks in SQL for filtering, reporting, completion tracking, and duplicate-task prevention.

Outcome

A working prototype tested on real email data, covering Gmail integration, priority detection, task extraction, daily briefings, and productivity analytics.

PythonSQLFastAPILangGraphLlama 3Gmail APIChromaDBStreamlit
View on GitHub

More Projects

Full-Stack May 2026 — Jul 2026

Enterprise HRMS Platform

Problem: managing the whole employee lifecycle in one secure system.

End-to-end HRMS covering employees, recruitment, attendance, payroll, leave, and performance, with an RBAC-secured FastAPI and PostgreSQL backend, a React and Vite frontend, and full Swagger docs.

FastAPIPostgreSQLReactJWT / RBAC
Backend Frontend
AI & GenAI Live Demo

DocLens

Problem: answers locked inside PDFs, scans, and images that can't be searched.

A document intelligence workspace that turns documents into structured summaries and grounded, page-cited answers, with Tesseract OCR for non-selectable text and a document-aware Q&A layer on top.

PythonOCRRAGStreamlit
Live Demo
AI & GenAI Personal Project

Hybrid RAG System

Problem: finding the most relevant passages across PDFs to ground answers.

A RAG pipeline built from scratch: dense retrieval with Sentence Transformers and ChromaDB, sparse retrieval with BM25, cross-encoder reranking, and Gemini for grounded answer generation.

PythonChromaDBBM25Gemini
View on GitHub
AI & GenAI Personal Project

AI Research Assistant

Problem: researching a topic across many web sources by hand.

A full-stack research agent that finds sources with Tavily, scrapes them with Firecrawl, and produces reports with an executive summary, findings, comparisons, and a confidence score, using a local Llama 3.2 model via Ollama.

Node.jsExpressTavilyOllama
View on GitHub
Research · In Progress Capstone, VIT

Quantum vs. Classical Federated Learning

Question: how much does uneven data across hospitals hurt federated learning, and do quantum classifiers hold up?

A team capstone comparing variational quantum classifiers against classical models on real UCI Heart Disease data. I built the federated learning infrastructure and experiment pipeline: the resumable, crash-safe training loop, the Dirichlet partitioner, and the runner behind all 5 experiment arms, including two unattended multi-hour overnight runs with zero failures.

PythonPennyLanePyTorchFederated Learningscikit-learn
View on GitHub
AI & GenAI · In Progress Research Phase

SecureChain AI

Problem: teams get long lists of dependency vulnerabilities with no way to decide what to fix first.

A dependency-risk assessment platform in active design, with a Dependency Trust Score, Composite Risk Score, and Remediation Priority Index to help teams prioritize supply-chain vulnerabilities.

ReactFastAPIPostgreSQLGemini
View on GitHub
Computer Vision TECDIA Internship

Jumbled Frame Reconstruction

Problem: restoring the correct order of a shuffled video.

Computes SSIM and histogram similarity between frames and uses optical-flow motion continuity to reassemble a coherent sequence.

PythonOpenCVSSIMOptical Flow
View on GitHub

Product Skills

Product Thinking, Technical Depth

Product & Analytics

Product RequirementsUser StoriesAcceptance CriteriaUser FlowsSuccess MetricsData AnalysisSQLPythonMicrosoft ExcelData VisualizationStreamlit

Product Execution

Agile / ScrumSDLCRequirements AnalysisTechnical DocumentationCross-functional CollaborationStakeholder CommunicationProblem SolvingQA & Testing

Design & Product Tools

FigmaMicrosoft PowerPointMicrosoft WordMicrosoft Excel

Technical

PythonSQLC++JavaJavaScriptTypeScriptFastAPIREST APIs

Databases

PostgreSQLSQLiteRelational DatabasesData ModellingSchema DesignQuery Optimization

AI & GenAI

LLMsRAGNLPPrompt EngineeringLangGraphVector SearchEvidence Grounding

Also built with: React, Next.js, Flask, GraphQL, ChromaDB, LangChain, Ollama, Docker, GitHub Actions, AWS, Git, Postman, JMeter, Pytest.

Certifications

Applied Learning

  • BCG GenAI Job Simulation

    Sep 2025

    Forage

    Built an AI-powered financial chatbot and practiced client-facing data analysis, applying GenAI to a business use case.

  • IBM Generative AI: Career Education Program

    Jul 2026

    IBM Developer Skills Network

    A structured foundation in generative AI concepts and applications, alongside hands-on builds like Cortex.

  • Tata Cybersecurity Job Simulation

    Jul 2025

    Forage

    Identity and access management analysis with audit and compliance considerations, relevant to access-control work like role-based permissions.

Leadership

Programs & Stakeholders

Point of Contact, HackVerse 2026

Sep 2026

National Cyber Security CTF · graVITas 2026, School of Computer Science and Engineering (SCOPE), VIT Vellore

Hours
24
Participants
80+
  • Ran point for a 24-hour national-level CTF with 80+ participants, coordinating faculty, sponsors, judges, guests, and event teams.
  • Owned sponsor-facing documentation and correspondence, and coordinated guest and judge requirements with the organizing team.
  • Managed run-sheets and execution dependencies, verified the scoreboard, resolved issues during the event, and supported event reporting.

Patent

Hardware-Aware Adaptive Intrusion Detection System and Method Based on Metaheuristic Feature Masking and Sparse Bayesian Classification

Co-inventor · Indian Patent Application No. 202641101767 · Publication No. IN202641101767 A1
Filed Aug 2026 · Published 28 Aug 2026 · Applicant: Vellore Institute of Technology

Get In Touch

Let's Build Something

Open to Software, Data, Product, and AI roles. Reach out directly or find me on GitHub and LinkedIn.

Email Me Resume LinkedIn GitHub

Noida, India · +91-9958841023