Muhammed
Shafi KT.

The Challenge

Transforming raw, unstructured datasets into executive-ready dashboards traditionally demands manual cleaning, SQL authoring, schema guesswork, and manual chart design.

Architectural Solution

Built a multi-stage analytics pipeline that ingests datasets, profiles distributions, validates schema integrity, and automatically generates interactive dashboards alongside a conversational AI assistant.

Key Engineering Features

• Built a multi-stage analytics pipeline for dataset ingestion, profiling, validation, and interactive dashboard generation
• Implemented DuckDB-based analytics querying large datasets directly from Cloudflare R2 without heavy disk download overhead
• Integrated Redis and BullMQ background processing to keep the API server responsive during heavy jobs
• Real-time job progress and status updates streamed directly to the frontend via Socket.IO
• Automated server-side PDF report generation using headless Puppeteer
• Natural-language Q&A and dataset interrogation powered by OpenAI SDK and LLM integration

Critical Technical Decisions

• DuckDB over Cloudflare R2: Executes columnar queries directly on remote objects, avoiding high memory footprint on the API server
• Redis + BullMQ Queue: Decouples resource-intensive dataset profiling and PDF compilation into isolated background workers
• Socket.IO WebSockets: Streams live ingestion percentages and step transitions without client polling overhead

Authentication & Deployment

Cookie-based JWT authentication with Google OAuth. Express backend deployed on AWS EC2 behind Nginx reverse proxy; React frontend deployed on Vercel.

The Challenge

Coordinating events across multiple stakeholder roles (venue managers, organizers, attendees, and admins) often creates fragmented workflows across ticketing, booking, validation, and attendee support.

Architectural Solution

Contributed to a production-grade multi-role event and venue platform built with Next.js, Node.js/Express, MongoDB, Docker, AWS S3, and an integrated RAG chatbot for real-time attendee queries.

Key Engineering Features

• Contributed to a multi-role platform (organizers, venue owners, attendees, admins) covering venue booking, ticketing, check-in, certificates, and a RAG-powered chatbot
• Built the Event module end to end: data model, Zod schema validation, service logic, controller, and REST routes
• Extended and customized the admin dashboard with complete event management and lifecycle controls
• Integrated JWT (httpOnly cookie) and Google OAuth authentication with organization-based role access control
• Dockerized the backend with multi-stage development and production images, integrated with CI/CD
• Connected AWS S3 for secure storage and delivery of event assets and attendee certificates

Critical Technical Decisions

• Organization-Based RBAC: Rigorous authorization matrix ensuring venue owners, organizers, attendees, and admins have strictly scoped permissions
• httpOnly Cookie JWT + Google OAuth: Eliminates token exposure in client-side storage, guarding against XSS while simplifying user onboarding
• Multi-Stage Docker Images: Separated development hot-reloading from lean, hardened production runtime images

Authentication & Deployment

JWT (httpOnly cookie) and Google OAuth with organization-based role access. Media and certificates stored in AWS S3. Backend fully Dockerized with CI/CD deployment pipelines.

The Challenge

Traditional keyword search fails on complex, multi-format knowledge bases, causing users to miss vital documentation or LLMs to hallucinate answers without context.

Architectural Solution

Engineered an end-to-end document-to-RAG pipeline using PostgreSQL and pgvector for high-dimensional vector storage, coordinated through a Python FastAPI AI microservice and Google Gemini LLM.

Key Engineering Features

• Built a document-to-RAG pipeline for text extraction, chunking, embedding generation, and vector storage using PostgreSQL/pgvector
• Developed a Python FastAPI AI microservice for embeddings and semantic retrieval, integrated with the Node.js backend
• Implemented vector similarity search using PostgreSQL's pgvector extension managed through Prisma ORM
• Engineered hybrid microservice communication between Next.js/Node.js application layer and Python AI services
• Integrated Google Gemini LLM for hallucination-resistant, context-grounded document summarization and Q&A

Critical Technical Decisions

• PostgreSQL + pgvector: Eliminates third-party hosted vector database lock-in by co-locating relational metadata and vector embeddings in PostgreSQL
• Python FastAPI AI Microservice: Isolates machine learning computations and embedding pipelines into a lightweight, high-performance async Python service
• Prisma ORM: Guarantees end-to-end type safety, reliable relational migrations, and clean schema governance

Authentication & Deployment

Hybrid architecture: TypeScript/Node.js backend interfacing with a Python FastAPI microservice. PostgreSQL database with pgvector extension.

The Challenge

Building a production e-commerce store requires robust inventory state management, reliable coupon logic, secure payment processing, and multi-tier role authorization without payment or cart discrepancies.

Architectural Solution

Architected a full-featured e-commerce platform featuring catalog management, persistent cart and wishlist collections, a flexible coupon engine, Razorpay payments, and a three-tier administrative dashboard.

Key Engineering Features

• Built an e-commerce platform with product management, cart, wishlist, coupons, role-based authorization, and admin dashboard
• Integrated JWT (httpOnly cookie) and Google OAuth authentication with role-based access control (superadmin, admin, user)
• Secure payment processing via Razorpay supporting card, UPI, and Cash-on-Delivery (COD) flows
• Cloudinary media management for optimized, responsive product image delivery and CDN caching
• Custom coupon engine supporting percentage and flat discounts, usage quotas, and expiration rules

Critical Technical Decisions

• Three-Tier RBAC: Provides strict authorization boundaries separating customers, administrators, and superadmins across all endpoints
• Decoupled Cart & Wishlist: Maintained as independent MongoDB collections with optimistic frontend synchronization for resilient checkout
• Cloudinary Image Pipeline: Automatic compression and thumbnail delivery offloading server bandwidth and improving mobile load times

Authentication & Deployment

JWT authentication with Google OAuth and role-based authorization. Frontend deployed on Vercel, Express REST API with MongoDB and Cloudinary.