🤖 AI-Native Portfolio

Built with v0 by Vercel & Next.js

This portfolio represents a shift in modern web development: Generative UI. Instead of writing boilerplate code from scratch, I collaborated with v0 to architect a high-fidelity interface, then refined the logic using the Next.js App Router.

---

⚡ The Workflow

This project was developed using a rapid "Prompt-to-Production" cycle:

1. Design: Natural language prompting in v0.dev to generate the core UI.
2. Refinement: Iterative component styling using Tailwind CSS and shadcn/ui.
3. Engineering: Integration of local assets and logic refinement in IntelliJ IDEA.
4. Deployment: Hosted on Vercel with a custom domain and automated CI/CD.

---

🛠️ Tech Stack

• UI Generation: v0.dev
• Framework: Next.js (App Router)
• Styling: Tailwind CSS
• Components: shadcn/ui
• AI Integration: Google Gemini API
• Data Source: football-data.org API
• Package Manager: pnpm
• Deployment: Vercel

---

⚽ Football AI Assistant

An autonomous, multi-agent sports intelligence platform. This system transcends traditional RAG (Retrieval-Augmented Generation) by implementing an autonomous decision-making layer that navigates between structured statistical schemas and unstructured live-web intelligence.

🔄 System Architecture

The assistant uses a decoupled agent strategy to ensure the UI remains snappy while the logic remains deep. This flow-state demonstrates how the Coordinator and Suggestion agents interact:

graph TD
    A[User Input] --> B{Coordinator Agent}
    B -->|Structured Stats| C[football-data.org API]
    B -->|News/Injuries/Rumors| D[Tavily Web Search]
    C --> E[Detailed Tactical Response]
    D --> E
    E --> F[Client-Side Hydration]
    F --> G[Background Suggestion Agent]
    G -->|Analyze Response| H[Generate Follow-up Question]
    H --> I[Update Ghost Text Placeholder]
    I --> J[Wait for TAB Key / One-Tap Pill]

    style B fill:#064e3b,stroke:#059669,stroke-width:2px,color:#fff
    style H fill:#064e3b,stroke:#059669,stroke-width:2px,color:#fff
    style C fill:#1e293b,stroke:#475569,color:#fff
    style D fill:#1e293b,stroke:#475569,color:#fff

Loading

🧠 Core Performance Heuristics

• Autonomous Tool Orchestration: Implements dynamic tool selection logic. The agent evaluates the query's temporal sensitivity—routing historical or competition-specific queries to a structured API, while diverting rumor, injury, or tactical shift inquiries to a real-time web-crawling layer.

• Predictive UX & Contextual Memory: * Vectorized Follow-ups: Instead of basic "Who is next?" prompts, the Suggestion Agent utilizes the previous response's context to generate advanced inquiries (e.g., analyzing xG overperformance or defensive transition vulnerabilities).

  • Heuristic Autocomplete: A "Ghost Text" implementation utilizing the append pattern from the Vercel AI SDK, allowing for zero-friction navigation through complex tactical data.

• Deterministic Persona Engineering: Utilizes a highly disciplined system prompt architecture to enforce clinical, analytical output directly from the model inference. This eliminates conversational noise and ensures the agent consistently maintains a high-density, professional tactical tone without requiring post-inference filtering.

• Multi-Model Resilience & Fallback: Configured with a model hierarchy (Gemini 2.0 Flash → 1.5 Pro) to ensure high availability and sophisticated reasoning even during peak API latency.

---

📊 Technical Implementation Highlights

• Logic Routing: Built using the Vercel AI SDK's streamText with maxSteps: 5 to allow for multi-stage tool calls and iterative reasoning loops.
• State Management: Intelligent syncing of local UI state (Ghost Text) with the AI SDK’s message history for seamless "Tab-to-Send" functionality and predictive placeholder updates.

---

🔍 Job Scout — AI Job Search Agent (Micro-Frontend)

A full-stack AI-powered job search agent deployed as a micro-frontend at bengredev.com/ai-lab/job-search-agent.

For full technical details, architecture, and implementation notes, see the dedicated repository: 👉 thushanthbengre22-dev/job-search-agent

---

🤖 Warren BotIt — Polymarket AI Paper Trading Bot

An AI-powered paper trading bot that scans live Polymarket prediction markets, uses Claude Sonnet to estimate probabilities, and sends trade signals via Telegram. Deployed as a proxied dashboard at bengredev.com/ai-lab/warren-bot-it.

For full technical details, architecture, and implementation notes, see the dedicated repository: 👉 thushanthbengre22-dev/warren-bot-it

---

✨ Features

• Responsive Design: Fully optimized for mobile, tablet, and desktop.
• Clean URLs: Modern routing with and without the # hash, optimized for SEO.
• Performance: High Core Web Vitals scores via Next.js Image and font optimization.
• Accessible UI: Built with Radix UI primitives via shadcn for full keyboard/screen-reader support.

---

🚀 Local Development

Since this project uses pnpm, you can get it running locally on your Mac in seconds:

1. Clone the repo:

   git clone [https://github.com/thushanthbengre22-dev/portfolio-react-app.git](https://github.com/thushanthbengre22-dev/portfolio-react-app.git)
   

2. Install dependencies:

   pnpm install
   

3. Set up Environment Variables: Create a .env.local file in the root directory and add your API keys:

   GOOGLE_API_KEY=your_gemini_api_key
   FOOTBALL_DATA_API_KEY=your_football_data_org_api_key

4. Start the development server:

   pnpm dev
   

5. View the site:

   Open http://localhost:3000 in your browser.

---

📬 Connect with Me

• Portfolio: bengredev.com
• LinkedIn: Thushanth Bengre
• GitHub: @thushanthbengre22-dev