Engineer / Builder / Problem Solver
Engineer • Builder • Architect • Innovator •Engineer • Builder • Architect • Innovator •
NO.02
Capabilities
Tech_Stack
001Spec_Sheet
Frontend
HTML5/CSS/JSTypescriptReactNext.jsSASSTailwind CSS
002Spec_Sheet
Backend
Node.jsExpress.jsPythonFastAPI
003Spec_Sheet
Databases
MongoDBCosmos DBPostgreSQLDynamoDB
004Spec_Sheet
Cloud/DevOps
AWSAzureJenkinsNginxBicepLinuxCI/CD Pipelines
005Spec_Sheet
AI/ML
Agentic AILangChainHuggingFaceMCPGenAIOpenAIGeminiAzure AIClaudeRAG / Agentic RAG
006Spec_Sheet
Tools
GitPostmanVS CodeCursorWindsurfAntigravityClaude CodeGitHub CopilotCodex
007Spec_Sheet
Other
REST APIsJWTWebSocketsJSON
PR
Featured Projects
NO.03
Selected Works
(27 Projects)
Work History
NO.04
Experience
5.5YRS XP
Aug 2024 – Present
KellyOCG
Senior AI Engineer | Full Stack Developer
GRACE Boost - AI-powered Browser Extension
Azure, Azure AD, OpenAI, TypeScript, Node.js, Cosmos DB, Preact, ESBuild
- Took a Chrome extension (Manifest V3) from concept to production in about a month. It embeds AI tools into recruiter workflows, adapts per-site automatically, and supports new domains without redeployment.
- The core AI layer runs on Azure OpenAI function calling — handles candidate summaries, ATS-optimized resume rewrites, branded job postings (7 templates), and a Bullhorn-specific agent that gathers requirements across multiple turns before returning structured output.
- Bullhorn's nested iframes were the hardest integration problem. Built a parsing engine and component scanner that reads form data, classifies UI element types, and reinjects AI content through simulated DOM events — this ended up being the only approach the host app would reliably accept.
- File uploads (PDF, DOCX, images) go through Azure Document Intelligence, with Worker Threads handling larger files. Auth sits on Azure AD OAuth 2.0 — RS256 JWT validation, automatic refresh on 401, and JWKS key caching with exponential backoff so token verification holds up during provider outages.
- Keeping content scripts and the Preact UI in sync meant running Zustand dual-access patterns across 4 Cosmos DB containers. Candidate search runs through Databricks SQL with AI-powered ranking, and the whole search UI lives inside a Shadow DOM to avoid style collisions with the host page.
- A telemetry layer captures token usage and tool traces, feeding PowerBI dashboards. Bundle stays under 300KB with ESBuild tree-shaking. CI/CD runs on Azure DevOps, infrastructure defined in Bicep/ARM.
Project GRACE 2.0 - GenAI Chat Application
Azure, Azure AD, OpenAI, Gemini, Claude, Next.js, Cosmos DB, Bicep, MCP, Agentic AI, RAG, LangChain
- One of the core developers on GRACE, an AI chat platform now used company-wide at Kelly Services. Stack is Next.js 14, React 18, TypeScript, and Tailwind CSS on Azure.
- The model routing layer dispatches conversations to GPT-4, Claude, Gemini, or Perplexity depending on context and persona settings. Also built and maintained 10+ specialized AI agents — each has its own system prompts, file processing, and extension hooks.
- Users upload documents to Blob Storage, which get indexed through Azure AI Search with Document Intelligence for OCR. When the AI answers a question from those documents, it pulls and cites the specific passages it used.
- Auth runs on NextAuth + Azure AD with role-based permissions and multi-tenant isolation. Built the admin dashboard that tracks usage metrics, model performance, and adoption. Achieved full Veracode compliance and led monthly security reviews off penetration test findings.
- Wrote a Cosmos DB migration framework from scratch — 40+ versioned migrations have gone through it, all with zero production downtime.
- Multimodal features cover DALL-E image generation, GPT-4 Vision for image analysis, and Azure Speech for TTS/STT. A mobile UX rework with dynamic textbox resizing hit 95% satisfaction in user feedback, along with a prompt library for saving common workflows.
- LangChain and MCP (over HTTP) extended the platform into agentic territory — file lookups, API calls, multi-step task execution. Also set up self-healing UI tests with the browser-use AI library that catch regressions before each deployment.
Apr 2023 – Jun 2024
Poumki Digital LLP
Full Stack & DevOps Engineer
Project Convergence - Data Analytics Platform
MERN Stack, AWS, Jenkins, Nexus, Linux, Nginx
- Set up the full CI/CD pipeline from scratch — self-hosted Jenkins on EC2, Nexus for artifacts, S3 for storage. Cut costs by about 30%. Also built a rollback pipeline that took failed releases from causing downtime to being a non-issue.
- Real-time features run on Socket.IO with REST APIs and JWT auth underneath. The API layer has 100% unit test coverage.
- MongoDB backups are automated through AWS Lambda for disaster recovery. Added Winston and LogLevel logging across the backend so production issues are actually visible when they happen.
- Large datasets were causing frontend slowdowns — fixed it with pagination and server-side search filters.
Project e-slide - Presentation Sharing Platform
React, SASS, DynamoDB, AWS Lambda, Amplify, GraphQL, PostgreSQL
- Built the presentation rendering engine in React with GraphQL — handles multi-format uploads and keeps state synchronized across viewers in real time.
- The backend runs serverless on AWS Amplify and Lambda, keeping operational overhead low.
- Added a moderation system on PostgreSQL: content flagging, role-based permissions, and tiered access control.
- International users were seeing slow load times, so moved static assets to CloudFront/S3 and optimized the bundle — 40% faster. Also added multilingual support with React-i18next.
Project Izycourse - Online Course Platform
React, SASS
- Built the student-facing frontend — progress tracking, interactive course modules, and enrollment flows.
- UI and accessibility work drove a 25% increase in mobile engagement and brought the app to WCAG compliance.
- Added multilingual support via React-i18next for students across multiple countries.
Mar 2022 – Apr 2023
Tata Consultancy Services (TCS)
AI/ML & Full Stack Developer - Rapid Labs
AI/ML Prototypes for Healthcare
AngularJS, Python, Django, Vue.js, Flask
- Built two healthcare AI prototypes on the Rapid Labs team — an NLP ticket classifier (85% accuracy) and a computer vision model for ulcer detection (92% precision).
- Created demo applications in Django + Vue.js/AngularJS so stakeholders could evaluate models in simulated clinical settings.
- Owned the full ML pipeline end-to-end: data preprocessing, augmentation, model training, deployment, and wiring everything into the frontend.
Jan 2021 – Mar 2022
Tata Consultancy Services (TCS)
Site Reliability Engineer
SRE Automation Tools
.NET, MySQL, C#
- Built the 'Core Analysis Tool' in C#/.NET — it automates root cause analysis on ServiceNow tickets and cut resolution time by about 50%.
- The 'Akamai Report Generator' and 'Auto Delegation Tool' automated data validation work that was consuming 500+ hours of manual effort per year.
- Improved monitoring and incident response processes, which measurably brought down MTTR across critical production environments.
Academic Background
NO.05
Education
Oct 2025 – Current
Master of Computer Applications (MCA) - AI/ML
Amrita Vishwa Vidyapeetham
Jul 2017 – Sep 2020
Bachelor of Computer Applications (BCA)
VLB Janakiammal College of Arts and Science, Coimbatore