Full-Stack Developer & AI Systems Builder
"I build systems that don't just work — they prove they work. From incident investigation engines with 89.5% measured accuracy to real-time CRDT collaborative studios, I engineer for correctness, not guesses."
Explore Systems • Interactive Terminal • Architecture & Telemetry • Tech Stack • Local Setup
This repository powers developer-portfolio-nu-rouge.vercel.app, a high-performance, dark-mode terminal-grade portfolio engineered with React 19, Tailwind CSS v4, and Vite.
Rather than presenting static resume bullet points, this platform functions as an empirical proof-of-work showcase featuring live architectural schematics, real-time hardware telemetry, and a grounded serverless AI terminal assistant powered by Gemini 2.5 Flash.
┌──────────────────────────────────────────────────────────────────────────────────────────────────────────────────┐
│ CORE SYSTEMS SUITE │
├───────────────┬───────────────────┬────────────────────────────┬─────────────────────────────────────────────────┤
│ SYSTEM │ CODENAME │ PRIMARY ARCHITECTURE │ VERIFIED BENCHMARK │
├───────────────┼───────────────────┼────────────────────────────┼─────────────────────────────────────────────────┤
│ SYS-01 │ TRACE │ Telemetry Critique Loop │ 89.5% vs 73.7% naive baseline │
│ SYS-02 │ CHRONOS │ Constraint Backtracking │ 46 nodes, 0 backtracks with MRV/LCV │
│ SYS-03 │ SyncPad │ Yjs CRDT + Pyodide WASM │ Conflict-free convergence via Yjs CRDT │
│ SYS-04 │ IncidentHub AI │ SRE Concurrency & OAuth │ 236/236 passing unit/integration tests │
└───────────────┴───────────────────┴────────────────────────────┴─────────────────────────────────────────────────┘
- Purpose: Evidence-Grounded Production Incident Investigation Engine investigating unseen production outages by executing deterministic scoring and an LLM critique loop instead of naive single-shot LLM guesses.
- Empirical Benchmark: 89.5% root-cause accuracy vs. 73.7% naive baseline across 19 synthetic incident scenarios with hidden root causes.
- Stack: Python, FastAPI, PostgreSQL,
pgvector, Gemini API, Next.js, TypeScript, D3.js, GSAP. - Key Components:
- Achieved 89.5% root-cause accuracy across 19 synthetic incidents with hidden root causes, versus a 73.7% single-prompt LLM baseline, using deterministic scoring and an LLM critique loop.
- Correlated logs, metrics, traces, and deployment events through temporal and causal evidence; added a deterministic trend-comparison check after failure-case analysis exposed an incorrect memory-leak attribution.
- Vector similarity telemetry retrieval with
pgvectorand PostgreSQL. - Interactive temporal DAG causal graph visualized with D3.js and smooth GSAP morphing.
- Purpose: MRV/LCV Constraint Solver with Live Search Visualization solving high-dimension Constraint Satisfaction Problems (CSP) for academic and organizational timetable scheduling.
- Empirical Benchmark: MRV/LCV found a valid 46-session schedule in 46 search nodes with zero backtracks; chronological search hit the 10-million-backtrack limit without finding a solution on the same input.
- Stack: React, Node.js, PostgreSQL, Gemini API, D3.js, Tailwind CSS.
- Key Components:
- Backtracking constraint solver with MRV and LCV heuristics demonstrating heuristic search by finding a valid schedule in 46 nodes with zero backtracks (chronological baseline hit the 10M-backtrack limit without finding a solution).
- Natural-language scheduling requirement translation into structured constraints via Gemini API.
- Side-by-side live D3.js visualization rendering search-tree traversal steps in real-time.
- Purpose: CRDT Collaboration with Sandboxed In-Browser Execution featuring conflict-free concurrent document synchronization and sandboxed execution.
- Empirical Benchmark: Conflict-free convergence via Yjs CRDTs + 100% client-side isolated runtime.
- Stack: React, TypeScript, Monaco Editor, Yjs, WebSockets, Web Workers, Pyodide (WebAssembly).
- Key Components:
- Enabled conflict-free concurrent editing with Yjs CRDT synchronization, live cursor sharing, and collaborator presence.
- Implemented browser-based execution for JavaScript, TypeScript through transpilation, and Python using Web Workers and Pyodide WebAssembly, keeping execution off the main UI thread.
- Purpose: Multi-tenant incident triage platform correlating telemetry across GitHub, Sentry, Slack, and Jira.
- Empirical Benchmark: 236/236 passing automated tests + 4 complete OAuth 2.0 flows.
- Stack: React 19, TypeScript, Node.js, Express, PostgreSQL, Redis, Socket.io.
- Key Components:
- Real OAuth 2.0 integrations with automated webhook ingestion pipelines.
- Distributed concurrency lock control with Redis Redlock to eliminate triage race conditions.
- Real-time incident war rooms powered by WebSockets.
The portfolio embeds an interactive terminal (rakesh@core:~$) directly connected to a serverless Gemini AI agent trained strictly on verified profile context.
┌─────────────────────────────────────────┐
│ USER TERMINAL PROMPT │
└────────────────────┬────────────────────┘
│
Normalized Cache Lookup?
/ \
YES / \ NO
/ \
┌──────────────────────┐ ┌───────────────────────────────┐
│ Return Instant Cache │ │ Invoke Gemini 2.5 Flash │
│ (In-Memory Cache Hit)│ │ (Strict Context + 8s Timeout) │
└──────────────────────┘ └───────────────┬───────────────┘
│
Response Arrived < 8.5s?
/ \
YES / \ TIMEOUT/ERR
/ \
┌──────────────────────┐ ┌───────────────────────────┐
│ Streamed Character │ │ Local Grounded Evidence │
│ Terminal Telemetry │ │ Fallback Notice │
└──────────────────────┘ └───────────────────────────┘
- Model Pipeline: Direct ultra-low-latency calls to
gemini-2.5-flash(~1.5s–2.5s response time). - Multi-Tier Response Cache: In-memory caching on client and edge ensures standard suggestion chips ("What makes TRACE unique?", "Why did you build CHRONOS?", "What's your tech stack?") return with near-instant in-memory cache hit.
- Zero Hallucination Guardrails: Programmed to answer exclusively using verifiable facts from
profile-context.json. Out-of-scope inquiries trigger transparent disclosures. - Circuit Breaker: An 8.5-second
AbortControllertimeout prevents indefinite hangs, gracefully triggering local evidence grounding if network latency spikes. - Interactive Controls: Supports standard bash commands (
whoami,projects --list,skills --list,experience --show,contact,help,clear).
-
Problem: Standard
(frameCount * 1000) / elapsedcounters drop to misleading values likeFPS 1during initial component mounting or when tabs are backgrounded. -
Solution: Implemented a 18-frame rolling ring buffer using
performance.now()andrequestAnimationFrame. -
Spike Filtering: Discards long frame intervals (
$\Delta t > 120\text{ms}$ ) caused by React mounting or GC pauses so lag spikes do not pull down the rolling average. -
Believable Window: Restricts numeric readout strictly between 30–240 FPS. If throttled or backgrounded, it gracefully renders
LIVEvia the Page Visibility API (visibilitychange), preserving layout symmetry without DOM shifts.
- Zero-Lag Calculation: Avoids layout thrashing by executing
getBoundingClientRect()within a throttledrequestAnimationFrameloop synced with Lenis. - Mathematical Precision: A section is active if and only if
rect.top <= focalY && rect.bottom > focalYwherefocalY = window.innerHeight * 0.35. Eliminates the classicIntersectionObserverpartial-entry jumping bug. - Unified Subsystems: Seamlessly maps both
#architectand#matrixto the$architectnavigation highlight.
- Configured
html.lenis, html.lenis-scrolling { scroll-behavior: auto !important; }to eliminate fighting between CSS smooth scrolling and JavaScript micro-stepping. - Calibrated wheel physics (
duration: 1.0,wheelMultiplier: 1.0) for 1:1 physical response on desktop mouse wheels. - Capped long navbar jump durations at 0.7s (700ms) for brisk, snappy navigation.
- Intro startup animation runs only on first arrival per browser session (
sessionStorage.getItem('bootComplete')). Subsequent navigations, section hashes (#systems,#matrix), and page refreshes render instantly.
| Category | Primary Technologies & Frameworks | Supporting Stack & Tools |
|---|---|---|
| Languages | JavaScript (ES6+), TypeScript, Python |
C++, SQL |
| Backend & APIs | Node.js, Express.js, FastAPI, REST APIs |
WebSockets, JWT, RBAC |
| Frontend | React.js, Next.js, D3.js, Tailwind CSS |
HTML5, CSS3 |
| Databases & ORM | PostgreSQL, MongoDB, MySQL, Redis |
pgvector, Prisma, SQLAlchemy |
| AI / LLM | Gemini API, Vector Embeddings, LLM Evaluation |
— |
| Systems | Yjs (CRDT), WebAssembly, Web Workers |
— |
| Tools, Testing & Deployment | Git, GitHub, Docker, Postman |
Vitest, Pytest, Supertest, Vercel, Render |
Developer-Portfolio/
├── api/
│ ├── ask.js # Serverless Gemini 2.5 Flash agent endpoint with caching & timeout
│ └── profile-context.json # Ground truth profile data for AI evaluation
├── public/
│ ├── Rakesh_Kumar_Resume.pdf# Production downloadable resume
│ └── favicon.svg # Cybernetic terminal favicon
├── src/
│ ├── components/
│ │ ├── About.jsx # Bio & quick-facts overview
│ │ ├── Architect.jsx # Operator spec sheet, 7-box capability matrix & certifications
│ │ ├── Contact.jsx # Terminal comms portal with copy-to-clipboard telemetry
│ │ ├── Footer.jsx # Symmetrical bottom bracket & social anchors
│ │ ├── Hero.jsx # Terminal headline, developer.json card & live CTAs
│ │ ├── HudFrame.jsx # Fixed corner brackets, system load & rolling FPS counter
│ │ ├── InteractiveTerminal.jsx # Bash emulator with grounded AI streaming & suggestion chips
│ │ ├── Navbar.jsx # Sticky header with focal-line scroll-spy indicator
│ │ ├── OpenSource.jsx # 6-card GitHub open-source signal grid
│ │ ├── Principles.jsx # Engineering tenets (Hypothesis Testing, Determinism, Correctness)
│ │ ├── Projects.jsx # 4 Deep-dive flagship systems with live architecture schematics
│ │ └── SystemBootloader.jsx# Session-persistent startup sequence
│ ├── data/
│ │ ├── portfolioData.js # Portfolio content constants & system specifications
│ │ └── profile-context.json # Client-side ground-truth profile mirror
│ ├── App.jsx # Application root container
│ ├── main.jsx # Entrypoint, Lenis smooth scroll initialization & routing
│ └── styles.css # Tailwind CSS v4 imports, custom scrollbar & Lenis styling
├── index.html # Semantic HTML5 template, Open Graph, Twitter & JSON-LD metadata
├── package.json # Dependencies & build scripts
├── vite.config.js # Vite configuration with local /api/ask proxy
└── README.md # Complete technical documentation
- Node.js:
v18.xorv20.x+(Recommended:v20.18.x) - Package Manager:
npm(v9+) orpnpm
git clone https://github.com/rakeshkumar0804/Developer-Portfolio.git
cd Developer-Portfolionpm installCreate a .env file in the root directory:
# Required for Interactive Terminal AI Agent (/api/ask)
GEMINI_API_KEY=your_google_gemini_api_key_hereTip: Obtain a Gemini API key from Google AI Studio. The application also includes offline local evidence fallback if no key is configured.
npm run devThe application will be accessible at http://localhost:5173.
npm run buildBuild output will be bundled into the dist/ directory.
npm run previewThe project is pre-configured for zero-configuration deployment on Vercel:
- Build Command:
npm run build - Output Directory:
dist - Serverless Function:
api/ask.jsautomatically deployed as an edge-ready serverless function. - Set
GEMINI_API_KEYin the Vercel Project Environment Variables.
Rakesh Kumar
Full-Stack Developer & AI Systems Builder
Gurugram, Haryana, India
- 🌐 Live Portfolio: developer-portfolio-nu-rouge.vercel.app
- 💼 LinkedIn: linkedin.com/in/rakesh-kumar-520754246
- 🐙 GitHub: @rakeshkumar0804
- 🧩 LeetCode: leetcode.com/u/Rakesh__Kumar_ (175+ Solved)
- ✉️ Email: rakeshchauhan6651@gmail.com
This project is licensed under the MIT License.