Edison Dev Portfolio & CyberStack AI

Welcome to the official development repository for Edison Isaza, a Senior Software Engineer specializing in Full Stack Development, Cybersecurity, and Financial Algorithms.

This project features CyberStack, an advanced AI Assistant designed to represent Edison professionally. It uses a custom "Neural Activation" architecture to allow the AI to act as an expert consultant, dynamically pulling context from LinkedIn and GitHub.

🚀 Key Features

🤖 CyberStack AI Agent

• Expert Persona: A specialized AI assistant that acts as a third-party expert representing Edison. It answers questions about his skills, experience, and services with high precision.
• Neural Activation: Powered by src/app/utils/expert_persona.xml, ensuring strict adherence to the expert persona and minimizing hallucinations.
• Dynamic Context:
  • GitHub: Fetches real-time project data from editech-dev/editech-dev.
  • LinkedIn: Injects detailed verified experience from src/data/profile.json.
• Chat Memory: Full conversation history persistence using Redis.
• AI Fallback: If Gemini fails, the chat API automatically falls back to OpenRouter (with the same Gemini Flash model).
• Dual AI Mode: Users can switch between Cloud AI (Gemini Flash) and Local AI (Chrome's built-in Gemini Nano) directly in the chat widget.
• Tool Calling: The agent can execute function calls to fetch profile data, repository lists, README content, and page view counts in real time.

📄 Interactive CV / Resume

• Bilingual: Toggle between Spanish and English with smooth Framer Motion transitions.
• Export: Print-to-PDF or download a professionally designed two-column PDF generated server-side via @react-pdf/renderer (/api/cv-pdf?lang=es|en).
• ATS-Optimized: Structured data from src/data/cv-profile.json ensures compatibility with automated resume screening systems.

🌐 Bilingual Repository System

• AI-Powered Translation: Repository descriptions and READMEs are automatically translated between Spanish and English using Gemini Flash Lite.
• Multi-Layer Caching: Translated content is cached in Redis (30-day TTL) and backed by a static fallback file (src/data/bilingual-repos-fallback.json).
• Change Detection: The system detects content alterations and re-translates only when the original text has changed.

📊 Real-time Analytics

• View Counter: Each project page tracks unique visitors via cookie-based UUID, stored in Redis with deduplication.
• Cache Revalidation: View increments trigger revalidatePath to keep the repositories list and detail pages up to date.
• Compact Display: View counts are formatted with Intl.NumberFormat compact notation (e.g., 1.2K).

⚡ Tech Stack

• Frontend: Next.js 16 (App Router), React 19, Tailwind CSS v4, Framer Motion.
• Styling: Tailwind CSS v4 with custom CSS modules, Geist Sans/Mono fonts, forced dark mode.
• AI Core: Google Gemini Flash Lite (using Google Generative AI SDK @google/genai). Optional fallback via OpenRouter.
• Database & Cache: Redis (chat logs, GitHub API cache, view counters, CV profile cache) with automatic in-memory fallback.
• PDF Generation: @react-pdf/renderer for server-side professional CV/resume PDF.
• Markdown Rendering: react-markdown + remark-gfm for chat message formatting.
• SEO: Dynamic Sitemap (sitemap.ts), Robots.txt (robots.ts), JSON-LD Person Schema, Open Graph metadata, canonical URLs.
• Testing: Vitest + React Testing Library + jsdom.
• Language: TypeScript 5.
• Package Manager: pnpm 11.12.0 (pinned via Corepack).

🗺️ Page Routes

| Route | Description | |---|---| | / | Home page with Neural Activation particles background, name title, and navigation | | /cv | Interactive bilingual CV/Resume with toggle and PDF export/download | | /contact | Contact page with LinkedIn, Email, and GitHub cards | | /repositories | List of public GitHub repositories with descriptions, stars, languages, and view counts | | /repositories/[repoName] | Repository detail page with README, bilingual toggle, and live view counter |

🛠️ Architecture

1. The Expert Brain (src/app/utils/expert_persona.xml)

The core of the agent is defined in an XML file that mimics "neural instructions". This file dictates:

• Identity: "Professional Assistant for Edison Isaza".
• Prime Directive: Always speak in the third person ("Edison has...", "He is...").
• Context Slots: Placeholders {{PROFILE_CONTEXT}} and {{GITHUB_CONTEXT}} are injected at runtime.

2. Context API (src/app/api/agent-context/route.ts)

This endpoint constructs the system prompt dynamically on every request:

1. Fetches public repos from GitHub.
2. Reads profile from Redis (with cv-profile.json local fallback).
3. Injects both into expert_persona.xml.
4. Returns the compiled "Brain" to the chat interface.

3. Chat API (src/app/api/chat/route.ts)

Handles streaming chat responses with function-calling capabilities:

• Primary: Google Gemini Flash Lite via @google/genai SDK with streaming, tool declarations, and automatic function call execution loops (max 5 iterations).
• Fallback: If Gemini fails (e.g., missing API key), automatically falls back to OpenRouter with SSE streaming.
• Tools: getProfileInfo, getGithubRepos, getRepositoryReadme, getProjectViews — defined in src/app/utils/chatbot_tools.ts.

4. Persistent Memory (src/app/api/log-chat/route.ts)

Every conversation turn is automatically logged to Redis.

• Log Chat: POST /api/log-chat
• List Chats: GET /api/chats (protected by ADMIN_API_TOKEN Bearer auth)
• View Chat: GET /api/chats/[id] (protected by ADMIN_API_TOKEN)

5. CV PDF Generation (src/app/api/cv-pdf/route.ts)

Generates a professionally designed two-column PDF on the server:

• Accepts ?lang=es|en query parameter.
• Reads profile from Redis/cv-profile.json, loads the photo from public/, and renders via @react-pdf/renderer.
• Returns PDF as downloadable attachment with language-appropriate filename.

6. Data Caching Layer (src/app/utils/github.ts)

To ensure speed and avoid GitHub API rate limits, the agent's knowledge of your projects is cached.

• Mechanism: src/app/utils/redis.ts handles generic caching with automatic in-memory Map fallback when Redis is unavailable.
• Cache Key: github:repos:public:bilingual stores the filtered, translated repository list (1-hour TTL).
• Bilingual Cache: Per-repo keys github:repo:bilingual:{name} store translated descriptions and READMEs (30-day TTL).
• README Filtering: Only repos with READMEs >= 100 characters are displayed.
• Translation Pipeline: src/app/utils/translation.ts uses Gemini Flash Lite for language detection and translation with exponential backoff retry (up to 3 retries for rate limits).

7. Real-time Analytics (src/app/api/views/[slug])

Tracks page views for each project repository with deduplicated unique visitor counting.

• Storage: Redis integer keys (views:{slug}) with Set-based dedup (viewed:{slug}) per user.
• Client: ViewCounter component uses cookie-persisted UUIDs (uniqueId) for visitor identification.
• Mechanism:
  • POST /api/views/[slug]: Increments view count and triggers Next.js revalidatePath.
  • GET /api/views/[slug]: Returns current view count.

8. Redis Client with In-Memory Fallback (src/app/utils/redis.ts)

The Redis client includes a complete in-memory mock layer that activates automatically when Redis is unreachable (2-second connection timeout). This ensures local development and preview deployments work seamlessly without a Redis instance.

📂 Project Structure

src/
├── app/
│   ├── api/
│   │   ├── agent-context/    # Dynamic context generation (XML + GitHub + Profile)
│   │   ├── chat/             # Streaming chat with Gemini + OpenRouter fallback + tool calling
│   │   ├── chats/            # Chat history retrieval (admin-protected)
│   │   │   └── [id]/         # Single chat detail
│   │   ├── cv-pdf/           # Professional PDF CV generation (@react-pdf/renderer)
│   │   ├── log-chat/         # Chat logging endpoint (to Redis)
│   │   └── views/
│   │       └── [slug]/       # View counter (POST increment + GET fetch)
│   ├── components/
│   │   ├── Card/             # Glassmorphic card with mouse-tracking gradient (Framer Motion)
│   │   ├── ChatBot/          # CyberStack chat widget (Cloud/Local AI toggle, streaming, markdown)
│   │   ├── Contact/          # Contact page with social link cards
│   │   ├── CvPage/           # Interactive bilingual CV + professional PDF component
│   │   ├── HomePage/         # Landing page layout
│   │   ├── MainPartiles/     # Home page hero with particle background
│   │   ├── Mdx/              # Custom MDX components for README rendering
│   │   ├── NavHomePage/      # Home page navigation bar (Projects, CV, Contact)
│   │   ├── Navigation/       # Global sticky nav with smart back button (IntersectionObserver)
│   │   ├── Particles/        # Canvas-based shell code rain particle effect
│   │   └── Repositories/     # Repository list, detail, bilingual toggle, view counter
│   ├── contact/              # /contact page
│   ├── cv/                   # /cv page
│   ├── repositories/         # /repositories page
│   │   └── [repoName]/       # /repositories/[repoName] dynamic detail page (SSG + ISR)
│   ├── utils/
│   │   ├── chatbot_tools.ts   # Gemini function-calling tool declarations and execution
│   │   ├── expert_persona.xml # Neural Activation System Prompt
│   │   ├── github.ts          # GitHub API Fetcher + Bilingual Translation Pipeline
│   │   ├── mouse.ts           # Mouse position tracking hook
│   │   ├── redis.ts           # Redis Client + In-Memory Fallback + View Counter
│   │   └── translation.ts     # AI-powered language detection + translation (Gemini)
│   ├── globals.css            # Forced dark theme, custom animations (glow, fade, title)
│   ├── layout.tsx             # Root layout with SEO metadata, JSON-LD, ChatBot, skip-to-content
│   ├── robots.ts              # Dynamic robots.txt generation
│   └── sitemap.ts             # Dynamic sitemap.xml with static + GitHub repository routes
└── data/
    ├── profile.json                  # Static Professional Profile (LinkedIn source)
    ├── cv-profile.json               # Detailed bilingual CV (bilingual sections)
    └── bilingual-repos-fallback.json  # Static translation fallback for repositories

🧪 Testing

The project uses Vitest with React Testing Library and jsdom for unit and integration tests.

pnpm test          # Run tests in watch mode
pnpm test:run      # Single run
pnpm test:coverage # With coverage report

Test files:

• src/app/api/chats/route.test.ts
• src/app/api/cv-pdf/route.test.ts
• src/app/components/Card/card.test.tsx
• src/app/components/ChatBot/ChatBotComponent.test.tsx
• src/app/components/CvPage/CvClientPage.test.tsx
• src/app/components/Repositories/ViewCounter.test.tsx

🤖 AI Agent Development Workflow

The project uses a multi-agent architecture for development via OpenCode, with 5 specialized subagents and 12 skill packs.

Subagents

| Agent | Role | Mode | |---|---|---| | ui-ux-designer | Tailwind CSS layouts, Framer Motion animations, responsive design | Read/Write | | a11y-auditor | WCAG 2.2 Level AA compliance, ARIA, keyboard nav, screen readers | Read/Write | | test-engineer | Vitest + React Testing Library unit/integration tests | Read/Write | | security-auditor | Dependency vulnerability auditing (pnpm audit) | ReadOnly | | git-manager | Git staging, Conventional Commits generation | Read/Write |

Skill Packs (12 installed)

| Skill | Source | Purpose | |---|---|---| | next-best-practices | vercel-labs/next-skills | RSC, App Router, data patterns, error handling | | react-best-practices | vercel-labs/agent-skills | Rendering optimization, bundle, server/client patterns | | composition-patterns | vercel-labs/agent-skills | Compound components, render props, context, variants | | next-cache-components | vercel-labs/next-skills | PPR, use cache directive, cacheLife, cacheTag | | tailwind-css-patterns | giuseppe-trisciuoglio/developer-kit | Responsive design, layout, spacing, typography | | frontend-design | anthropics/skills | Premium design & aesthetics | | typescript-advanced-types | wshobson/agents | Generics, conditional types, mapped types | | accessibility | addyosmani/web-quality-skills | WCAG 2.2 audit patterns | | seo | addyosmani/web-quality-skills | Meta tags, structured data, sitemap | | next-upgrade | vercel-labs/next-skills | Next.js version migration guides | | nodejs-best-practices | sickn33/antigravity-awesome-skills | Async patterns, security, architecture | | nodejs-backend-patterns | wshobson/agents | Express/Fastify middleware, error handling, auth |

📦 Getting Started

1. Clone the repository:

   git clone https://github.com/editech-dev/edison-dev.git
   cd edison-dev
   

2. Enable Corepack and install dependencies (pnpm only — the project pins [email protected] via the packageManager field):

   corepack enable
   pnpm install
   

3. Configure Environment: Create a .env.local file with:

   # Required
   GEMINI_API_KEY=your_gemini_key
   GITHUB_TOKEN=your_github_token
   GITHUB_REPO_OWNER=editech-dev
   
   # Optional
   REDIS_URL=redis://localhost:6379           # Falls back to in-memory store if missing
   OPENROUTER_API_KEY=your_openrouter_key     # Fallback if Gemini fails
   ADMIN_API_TOKEN=your_admin_token           # Protects /api/chats endpoints
   NEXT_PUBLIC_BASE_URL=https://editech.dev   # Used for sitemap, robots.txt, OG metadata
   

4. Run Development Server:

   pnpm dev
   

5. Access the App: Open http://localhost:3002. The CyberStack chat widget will be available globally.

📊 API Documentation

Chat & AI

| Endpoint | Method | Auth | Description | |---|---|---|---| | /api/chat | POST | None | Streaming chat with Gemini (primary) or OpenRouter (fallback). Body: { messages, systemInstruction } | | /api/agent-context | GET | None | Returns compiled system prompt with live GitHub data + CV profile injected | | /api/log-chat | POST | None | Persists chat session to Redis. Body: { messages, timestamp, chatId? } | | /api/chats | GET | Bearer ADMIN_API_TOKEN | Lists all chat sessions with metadata (id, timestamp, preview) | | /api/chats/{id} | GET | Bearer ADMIN_API_TOKEN | Retrieves a single chat session by ID |

CV / PDF

| Endpoint | Method | Description | |---|---|---| | /api/cv-pdf?lang=es\|en | GET | Generates and downloads a professionally designed two-column ATS-optimized PDF resume |

Analytics

| Endpoint | Method | Description | |---|---|---| | /api/views/{slug} | POST | Increments view count for a repository (deduplicated by cookie uniqueId). Body: { uniqueId }. Triggers Next.js revalidatePath | | /api/views/{slug} | GET | Returns current view count for a repository |

🚑 Troubleshooting

Vercel production build fails with ERR_PNPM_IGNORED_BUILDS

Symptom (observed 2026-07-19): Preview deployments built successfully, but Production deployments died during dependency installation, before next build even started:

[ERR_PNPM_IGNORED_BUILDS] Ignored build scripts: @google/[email protected], [email protected], [email protected], [email protected]
Run "pnpm approve-builds" to pick which dependencies should be allowed to run scripts.
Error: Command "pnpm install" exited with 1

Root cause — two factors combined:

1. pnpm 11 breaking changes. The repo pins packageManager: [email protected]. In pnpm 11, strictDepBuilds defaults to true: any dependency shipping lifecycle scripts (preinstall/postinstall/install) that has not been explicitly approved or denied makes pnpm install exit with code 1. pnpm 11 also removed the legacy settings (onlyBuiltDependencies, neverBuiltDependencies, ignoredBuiltDependencies, …) and no longer reads configuration from package.json (pnpm field) or .npmrc/.pnpmrc (those files are auth/registry-only now). Consequently, former fix attempts — a pnpm.onlyBuiltDependencies field, a .pnpmrc file, and a PNPM_ONLY_BUILT_DEPENDENCIES Vercel env var — were all silently ignored.
2. Vercel environment divergence. ENABLE_EXPERIMENTAL_COREPACK=1 was configured for the Production environment only, so Production resolved the pinned pnpm 11.12.0 through Corepack. Preview had no such variable, so Vercel fell back to pnpm 9.x, which executes dependency build scripts by default — hence "works in Preview, fails in Production".

Fix applied:

1. pnpm-workspace.yaml (repo root) explicitly approves the four dependencies that ship install scripts, satisfying strictDepBuilds:

   allowBuilds:
     "@google/genai": true
     esbuild: true
     protobufjs: true
     sharp: true
   

2. ENABLE_EXPERIMENTAL_COREPACK=1 enabled for all Vercel environments (Production, Preview, Development) so every deployment installs with the same pinned pnpm version.
3. Removed the dead PNPM_ONLY_BUILT_DEPENDENCIES env var from Vercel (ignored by pnpm 11).

Rejected alternatives:

• strictDepBuilds: false — the install passes, but the scripts still do not run, leaving sharp/esbuild potentially degraded at runtime.
• dangerouslyAllowAllBuilds: true — works, but disables pnpm 11's supply-chain protection entirely.
• Downgrading to pnpm 10 — loses the version pin and security posture for no real gain.

References: pnpm 11 release notes · pnpm allowBuilds setting · pnpm strictDepBuilds setting · Vercel Corepack support

📄 License

This project is licensed under the MIT License. You are free to use, modify, and distribute this software for your own projects. See the LICENSE file for more details.

Built with ❤️ by Edison Isaza