Portfolio Website: sreekarreddy.com | GitHub: @esreekarreddy
A portfolio of production-grade AI applications showcasing cloud AI orchestration, local-first inference, generative UI, and in-browser AI execution. Each project demonstrates advanced architectural patterns and modern web technologies.
1. SR Terminal
A browser-based operating system with an integrated AI assistant, built entirely client-side using WebContainers and WebLLM.
Tech Stack: Next.js 16, WebContainer API, WebLLM (MLC), XTerm.js, Monaco Editor, Zustand, Tailwind CSS v4
AI Model: Microsoft Phi-3-mini-4k-instruct (3.8B parameters, quantized to 4-bit) — runs entirely in-browser via WebGPU
Key Features:
- In-Browser Node.js Runtime — Full Node.js 18 execution without a backend server
- AI Code Assistant — Natural language to code generation with on-device inference
- Multi-Terminal Interface — Tabbed shell sessions with parallel process management
- Monaco Code Editor — VS Code-grade editing with syntax highlighting and IntelliSense
- Resource Monitor — Real-time VFS and heap memory tracking
- Aerospace UI — Cinematic dark theme with custom modals and smooth animations
2. Cortex
A multi-agent code review system where six specialized AI agents analyze code from different perspectives, producing ranked findings with actionable fixes.
Tech Stack: Next.js 16, TypeScript, Tailwind CSS v4, Zustand, Monaco Editor
AI Models (via Ollama Cloud):
| Agent | Role | Cloud Model |
|---|---|---|
| Moderator | Orchestration & Verdict | gpt-oss:120b |
| Architect | Structure & Patterns | gpt-oss:120b |
| Sentinel | Security & Bugs | deepseek-v3.1:671b |
| Optimizer | Performance | qwen3-coder:480b |
| Maintainer | Tests & Refactoring | devstral-2:123b |
| Verifier | Claim Validation | gpt-oss:120b |
Key Features:
- Multi-Agent Cloud Analysis — 6 specialized agents powered by Ollama Cloud
- Massive Models — Access to 120B-671B parameter models
- Robust Error Handling — Retry logic, rate limit recovery, agent failover
- Severity-Ranked Findings — Issues sorted P0 (Critical) to P3 (Minor)
- Access Protected — Secret code prevents API abuse
3. Mirage
A sketch-to-code engine that uses Vision AI to transform hand-drawn wireframes into production-ready React components with live preview.
Tech Stack: Next.js 16, tldraw, WebContainer API, Ollama Cloud, Tailwind CSS v4
AI Model: Qwen3-VL 235B (Ollama Cloud) — vision-language model that "sees" your sketch
Key Features:
- Vision-Powered Generation — AI analyzes your sketch image and generates matching code
- In-Browser Vite Runtime — Complete dev server running client-side with hot-reload
- Iterative Refinement — Chat with the AI to modify generated components
- 1-Hour Access Expiry — Secure access code system with automatic expiration
- Cyberpunk Aesthetic — Dark theme with violet/cyan ambient glows
4. SR Mesh
A local-first 3D knowledge graph that uses client-side AI to generate semantic embeddings and visualize note relationships in 3D space. Features intelligent text classification and enhanced tooltip visibility.
Tech Stack: Next.js 16, Transformers.js (WebAssembly), React Three Fiber, IndexedDB, Vitest
AI Model: all-MiniLM-L6-v2 (sentence-transformers) — runs entirely in-browser via WASM
Key Features:
- On-Device Embeddings — 100% privacy-focused inference using Web Workers
- 3D Force-Directed Graph — Interactive particle visualization with physics simulation
- Semantic Clustering — Notes grouped by meaning, not just keywords
- Intelligent Text Classification — Rule-based categorization into 9 categories (Questions, Insights, Facts, Learning, Projects, Personal, Work, Ideas, Creative)
- Enhanced Tooltips — Large, readable hover info at any zoom level
- IndexedDB Persistence — Offline-first storage with full-text search
| Capability | Implementation |
|---|---|
| Cloud AI Orchestration | Ollama Cloud API with multi-model agents |
| Vision AI | Qwen3-VL for sketch-to-code conversion |
| In-Browser AI | WebLLM (WebGPU), Transformers.js (WASM) |
| Browser-Based Node.js | WebContainers for sandboxed execution |
| State Management | Zustand with TypeScript |
| UI Framework | Next.js 16 App Router + Tailwind CSS v4 |
All projects are built with security in mind:
- API keys never exposed to client — Server-side environment variables only
- Access code protection — Prevents unauthorized API usage on live demos
- Local-first architecture — SR Terminal and SR Mesh run entirely client-side
- HTTP security headers — HSTS, X-Frame-Options, X-Content-Type-Options, Referrer-Policy
- Input validation — Prototype pollution protection in
safeJsonParse, path traversal prevention - HttpOnly cookies — Mirage uses secure cookie-based auth with 1-hour expiry
Built by Sreekar Reddy • GitHub