Problem: Most AI chatbots are: Hardcoded for a single application Not reusable across projects Expensive due to inefficient prompt design Lacking proper API-level abstraction Not production-ready for SaaS monetization There was no lightweight, context-aware, token-optimized AI chatbot system that developers could integrate easily into any product.

Contribution: Designed and built a universal AI chatbot architecture Implemented context-aware prompt generation engine Created API-based chatbot integration model Designed token-efficient prompt strategy Built modular structure ready for SaaS monetization Implemented scalable public API endpoint architecture Deployed production-ready version on Vercel

Impact: Reduced prompt token usage through optimized layering Enabled cross-project chatbot reusability Created a foundation for API-based AI monetization Designed scalable architecture for future SaaS expansion Demonstrated advanced AI system design capability

Problem: Learners in Sri Lanka and emerging markets lack a polished, locally-contextualised e-learning platform that addresses local pricing (LKR), features high-quality UX, and supports flexible course discovery with rich filtering tools — most global platforms are too expensive or inaccessible.

Contribution: Built the entire public-facing client from scratch — landing page (hero, stats, features, how-it-works, testimonials, instructor carousel, CTA, footer), course listing with sidebar filtering and search, individual course detail pages, auth flows (login/signup), pricing, checkout, contact, about, and supporting pages (privacy policy, terms).

Impact: Delivered a production-ready SaaS-style learning frontend deployed on Vercel Implemented a premium dark glassmorphism design system with micro-animations Supports 50K+ active learners scale with optimised TanStack Query data fetching Built with SEO best practices (sitemap, robots.txt, metadata per page)

Problem: Taxi mobile apps often lack a professional marketing website to communicate service value, features, and encourage downloads or user engagement.

Contribution: • Designed and developed responsive marketing website • Built clear UI sections for features, testimonials, and call‑to‑action • Optimized for performance and SEO • Deployed on Vercel for reliable hosting

Impact: • Improved mobile app visibility and user engagement • Provided clear service information and call‑to‑action • Demonstrated frontend design and web‑marketing skills

Problem: Wedding planning involves coordinating multiple vendors, bookings, and schedules, often using disconnected tools that reduce efficiency and increase complexity.

Contribution: • Designed and developed full-stack wedding planning SaaS platform • Built vendor listing, booking, and management system • Developed enterprise admin dashboard for platform management • Implemented secure authentication and scalable backend • Created modular and production-ready architecture

Impact: • Enabled centralized wedding planning and vendor management • Improved efficiency of vendor booking and event coordination • Demonstrated enterprise-level SaaS system architecture skills • Built scalable platform suitable for real-world deployment

Problem: Hotel management operations often rely on manual tracking or fragmented systems, leading to inefficiencies, booking errors, and poor operational visibility.

Contribution: • Designed and developed full-stack hotel management system • Built booking and reservation management functionality • Implemented secure authentication and admin dashboard • Created scalable backend with REST API architecture • Developed responsive frontend using React and TypeScript

Impact: • Automated hotel booking and management workflows • Improved operational efficiency and booking accuracy • Enabled centralized management of rooms, guests, and reservations • Demonstrated full-stack SaaS system design capability

Problem: Most AI coding assistants rely on cloud APIs, lack customization, and do not provide persistent memory, offline capability, or full integration with local development workflows.

Contribution: • Developed full VS Code extension powered by custom fine-tuned AI model • Built AI orchestrator with model routing, prompt engine, and tool execution • Implemented local AI server using Ollama and llama.cpp • Designed memory engine using SQLite and vector embeddings • Integrated Figma plugin for AI-assisted UI generation and design automation

Impact: • Enabled fully local AI-powered development without external API dependency • Improved developer productivity with intelligent code and UI assistance • Created scalable AI development ecosystem with persistent memory • Demonstrated expertise in LLM integration, developer tools, and system architecture

Problem: Manual lead collection from platforms like Google Maps is time-consuming, inefficient, and difficult to scale for business development and outreach workflows.

Contribution: • Developed automated Google Maps scraping engine using Python • Built full-stack SaaS platform with FastAPI backend and React frontend • Implemented JWT authentication, multi-tenant architecture, and lead management system • Created scalable infrastructure with Docker, PostgreSQL, and Redis • Integrated analytics, monitoring, and lead scoring features

Impact: • Automated lead generation workflow and eliminated manual data collection • Created scalable SaaS platform for business intelligence and outreach • Improved efficiency and accuracy of business data collection • Demonstrated expertise in scraping, SaaS architecture, and full-stack development

Problem: Traditional classified platforms rely heavily on manual categorization and basic search, resulting in poor relevance, inefficient ad discovery, and limited automation.

Contribution: • Designed and developed full-stack AI-powered classified ads platform • Implemented intelligent ad categorization and search functionality • Built scalable frontend and backend architecture • Developed responsive user interface and ad management system • Integrated cloud deployment and database support

Impact: • Improved ad discovery and relevance using intelligent automation • Created scalable and modern classified ad platform • Demonstrated AI integration in real-world web applications • Showcased full-stack and AI-powered system development skills

Problem: Many individuals and teams lack simple, scalable, and cloud-based tools to efficiently manage tasks, workflows, and productivity in a centralized system.

Contribution: • Designed and developed full-stack SaaS task management platform • Built responsive frontend using Next.js and TypeScript • Integrated Firestore database for real-time data storage • Implemented secure authentication and user session management • Deployed scalable cloud application using Vercel

Impact: • Created modern productivity platform for task and workflow management • Demonstrated SaaS application architecture and deployment skills • Enabled scalable and cloud-based productivity system • Showcased full-stack web development and system design expertise

Problem: Many educational platforms lack scalable, customizable, and cost-efficient LMS solutions that support modern SaaS deployment and multi-user environments.

Contribution: • Designed and developed a full-stack SaaS learning management platform • Implemented authentication, course management, and user dashboards • Built scalable backend architecture supporting multi-user access • Developed responsive frontend using modern web frameworks • Integrated database for persistent course and user data

Impact: • Created scalable SaaS-ready LMS platform • Enabled structured digital learning and course delivery • Demonstrated full-stack SaaS architecture skills • Built production-ready cloud-deployable application

Problem: Enterprises and teams need a streamlined, all-in-one platform that accurately interprets natural language intents to build, execute, and monitor complex automated workflows without juggling separated tools and disjointed third-party platforms.

Contribution: Built a unified, multi-tenant monorepo architecture separating a high-performance Next.js application and a scalable Node.js/Express execution engine. Integrated semantic AI search with Xenova transformers, a robust Stripe checkout and billing layer with tier enforcement, and comprehensive workspace isolation (RBAC).

Impact: Provides businesses with a secure, highly-reliable automation workflow engine (comparable to n8n) embedded directly into a modern SaaS lifecycle. It empowers users to construct intelligent process topologies locally, scale them via Redis and BullMQ, and monitor their status efficiently with built-in audit logging and security protections.

Problem: Pixel LED hardware systems require specialized software to design, simulate, and upload visual patterns efficiently. Existing tools lacked integrated simulation, licensing control, and streamlined upload workflows.

Contribution: • Developed full desktop application for LED pattern design and upload • Built simulation engine to preview hardware behavior before deployment • Integrated secure licensing and subscription backend system • Implemented WiFi-based upload system for real-time hardware communication • Created scalable architecture combining desktop and web backend

Impact: • Enabled efficient design and deployment of LED matrix visual patterns • Improved development workflow through simulation and preview tools • Created production-ready hardware integration platform • Demonstrated expertise in desktop, backend, and hardware-integrated systems

Problem: Generic LLMs lack domain-specific intelligence and cannot reliably perform specialized engineering, system monitoring, automation, and bilingual operational tasks. They also lack real-time system integration and automation capabilities required for productivity workflows.

Contribution: • Fine-tuned Mistral-7B using LoRA with 137,300 structured training examples • Designed a multi-mode AI orchestration system with 7 operational modes • Built full AI assistant architecture including training, inference, deployment, and automation • Developed CLI interface, cloud deployment integration, and local inference pipeline • Implemented bilingual English and Sinhala support • Integrated system monitoring, business automation, and engineering assistant capabilities

Impact: • Created a production-ready custom AI assistant with 95-99% domain-specific accuracy • Reduced reliance on generic LLM APIs by enabling local and cloud inference • Enabled automation of engineering, monitoring, design, and business workflows • Built scalable AI architecture supporting cloud, desktop, and local deployments • Demonstrated advanced LLM fine-tuning, deployment, and orchestration skills

Problem: E-commerce businesses lacked affordable, integrated AI recommendations and business intelligence tools. They struggled with complex text-based ordering (like WhatsApp), multi-tenant data isolation, and scattered payment gateways.

Contribution: Architected and built a production-ready, multi-channel SaaS platform with AI-powered recommendations, natural language order parsing, and an integrated real-time analytics engine.

Impact: Delivered a scalable backend, integrated payments, and AI-driven insights for store owners, supporting automation across inventory, financials, and customer support.