Sarah Zephyr
Open to opportunities

Mesa SarahZephyr.

Software engineer specialising in full-stack, backend and AI/ML systems — I design, build and ship products end-to-end.

01 · About

Software engineer building full-stack, backend & AI systems.

Software engineer with hands-on experience across full-stack development, backend engineering and applied AI/ML. I enjoy shipping products end-to-end using Python, TypeScript, React, FastAPI, Node and modern ML tooling — turning ideas into deployed systems that serve real users.

Recent work spans building enterprise document-intelligence platforms at TCS, an offline RAG system for DRDO researchers, and shipping personal products like Nova, GraphMind AI and ChessMeet India.

Education

B.Tech in Computer Science Engineering

Vellore Institute of Technology, Amaravati

Relevant Coursework
  • Data Structures & Algorithms
  • System Design
  • Machine Learning
  • Natural Language Processing
Full-stack fluent

React, TypeScript, FastAPI, Python, SQL and Neo4j — schema to UI.

Production instincts

CI/CD, Docker, 85%+ test coverage, air-gapped deployments, real-time monitoring.

Available worldwide

Remote-first. Open to full-time, contract and interesting collaborations.

02

Experience.

Generative AI Intern

Tata Consultancy Services (TCS)

Feb 2025 – Apr 2025
Hyderabad Feb 2025 – Apr 2025
  • Built AI solutions for document processing, semantic search and knowledge retrieval for enterprise clients
  • Developed and deployed 6+ production REST APIs with 85%+ test coverage and CI/CD integration
  • Optimized database performance — 70% retrieval latency reduction across 10,000+ enterprise documents
  • Integrated Neo4j knowledge graph storage and Apache Solr for advanced semantic search
  • Worked in Agile / JIRA sprint cycles with peer code reviews
  • Docker-based deployment across staging-to-production pipelines
Project Intern

Defence Research & Development Organisation (DRDO)

Sep 2023 – Nov 2023
Hyderabad Sep 2023 – Nov 2023
  • Built an offline RAG system for 100+ R&D users to search technical documents without internet
  • Full pipeline: PDF ingestion → Tesseract OCR → embeddings → ChromaDB → LLM response
  • Improved OCR accuracy from 60% to 85–90% through preprocessing and chunking
  • Handled 1–2 GB document collections with efficient retrieval systems
  • Production deployment in a secure, air-gapped environment with zero downtime
  • Real-time troubleshooting, performance monitoring and maintenance for critical R&D workflows
03 · Selected Work

Projects that went from idea to production.

Tap a chip to filter by stack. Every card is a real project — deployed, tested, and used by real people.

Atlas AI

Autonomous AI Agent Orchestration Platform

Enterprise AI agent orchestration platform that transforms natural language goals into executable workflows using dependency graphs, parallel execution and human approvals.

  • Autonomous workflow planning powered by LLMs
  • Dependency Graph (DAG) execution with parallel task scheduling
  • Human approval checkpoints before high-risk actions
  • Real-time execution timeline via WebSockets
  • Sandboxed Python execution with retry & recovery
Next.jsFastAPIPostgreSQLDockerWebSocketsLlama 3Groq

LoadGhost

AI Generated Load Testing Platform

AI-powered load testing platform that converts real production traffic into executable performance tests with intelligent user journey discovery.

  • Automatic user journey discovery from traffic logs
  • Generates k6, Locust, JMeter and Artillery scripts
  • Interactive flow visualization and AI performance insights
  • Background task processing using Celery and Redis
  • Supports Nginx, HAR, GA, Mixpanel and OpenTelemetry
Next.jsFastAPIPostgreSQLRedisCeleryDockerPython

Magenta Mind

Enterprise Multimodal RAG Platform

Enterprise multimodal Retrieval-Augmented Generation platform supporting intelligent search across documents, videos and audio with grounded AI responses.

  • Multimodal RAG over documents, audio and video
  • Semantic vector search using pgvector
  • Timestamped AI citations with contextual evidence
  • OCR, transcription and embedding pipelines
  • Enterprise-ready scalable architecture
ReactTypeScriptPostgreSQLpgvectorOpenAITanStack
Code Demo on request

ChessMeet India

Live Video Chess Platform

Online chess platform with live multiplayer gameplay and peer-to-peer video calling, tailored for the Indian market.

  • Legal chess moves, checkmate detection and en passant with real-time sync via Socket.IO
  • Scalable TypeScript + React 18 + Express architecture designed for future expansion
  • WebRTC peer-to-peer video communication with safety-first features and dark mode UI
ReactTypeScriptViteSocket.IOWebRTCExpressPrisma
Private repo Demo on request
04 · Toolkit

The stack I reach for.

Languages, frameworks and infrastructure I've used in production or serious personal projects.

Languages

01
PythonJavaJavaScriptTypeScriptSQLC++Bash

Frontend

02
React.jsNext.jsHTMLCSSResponsive Design

Backend

03
FastAPIFlaskNode.jsExpressREST APIs

AI & Machine Learning

04
LangChainRAGLLMsTensorFlowPyTorchScikit-learnNLPComputer VisionPrompt Engineering

Databases

05
PostgreSQLMySQLMongoDBNeo4jChromaDB

Tools & DevOps

06
DockerGitGitHubLinuxCI/CDJira
05 · Certifications

Verified learning.

HarvardX

CS50's Introduction to Computer Science

HarvardX

CS50's Web Programming with Python and JavaScript

IBM

IBM Full Stack Software Developer Professional Certificate

Google

Google Data Analytics Professional Certificate

06 · Publications

Research on the record.

Conference paperNovember 2025

Research paper — accepted for presentation & publication

37th National Convention of Marine Engineers & National Seminar
The Institution of Engineers (India) (IEI), Andhra Pradesh State Centre

Presented work before engineers, researchers and industry professionals at a national-level technical conference.

07 · Leadership & Awards

Beyond the codebase.

Hackathons, community-building and curiosity that started well before college.

2026

Meta Hackathon Finalist

Developed and presented a working prototype under tight competition deadlines.

2024 – 2025

Machine Learning Club Manager, VIT-AP

Organized 10+ AI/ML workshops and grew club membership by 60%.

2024

MATHack 2.0 Finalist

Built an AI-powered NLP solution for talent screening, presented at the Global AI Summit.

2017

NASA Space Settlement Contest

Internationally selected project for innovative space colony design.

08 · Contact

Let's build something wonderful.

Open to full-time roles, internships and interesting AI / full-stack collaborations.

Say hi — I read every message.

Fastest way is email, but LinkedIn works too. Based in Available worldwide · Remote, working IST.

Send message