Portfolio AI

A conversational portfolio. Instead of a static page, visitors ask questions and an LLM answers them from a structured biography — behind a signature "shuffle" that loads one of 23 hand-designed art-style pages on every visit.

It's a conversational layer over a living professional profile: a knowledge base that's continuously crafted and structured together with AI and evolves over time, rather than a one-off CV or LinkedIn export.

How it works

server.js            Express: serves public/, streams /api/ask over SSE
lib/llm.js           Swappable LLM provider (local Ollama; hosted OpenCode Go)
content/             The biography the AI answers from — git-ignored, never served
public/              The entire web surface
  index.html         Manifest-driven shuffle launcher
  options.html       Gallery of every style
  llm-hook.js        Shared chat controller injected into every style
  options/
    manifest.json    Single source of truth (styles, captions, chips, contact, about)
    option-*.html    23 art-style pages

A hardened, single-purpose system prompt keeps the assistant scoped to questions about the subject only — it is not a general-purpose chatbot and refuses off-topic use and prompt-injection attempts. Requests are rate-limited and length-capped.

Privacy model

The biography is private and is never committed or served:

• The real bio lives at content/private-context.md, which is git-ignored.
• Only the sanitized content/private-context.example.md template is tracked.
• The bio sits outside the public/ web root, so it can't be served as a static asset under any route.
• Visitor questions are logged to logs/ (git-ignored) without IP or PII.

Run locally

Requires Node.js. The model runs either locally via Ollama (default) or on a hosted, OpenAI-compatible gateway (OpenCode Go) — see Configuration below. npm start auto-loads a .env file if present.

npm install

# Pull the model (default: gemma4:12b-mlx — override with OLLAMA_MODEL)
ollama pull gemma4:12b-mlx

# Create your bio from the template
cp content/private-context.example.md content/private-context.md
#   …then edit content/private-context.md with your real biography

npm start
# → http://localhost:8787

Configuration

Variable	Default	Purpose
PORT	8787	HTTP port
LLM_PROVIDER	ollama	ollama (local) or opencode (hosted)
OLLAMA_URL	http://127.0.0.1:11434	Ollama endpoint
OLLAMA_MODEL	gemma4:12b-mlx	Local model
OPENCODE_API_KEY	—	OpenCode Go key (https://opencode.ai/auth)
OPENCODE_BASE_URL	https://opencode.ai/zen/go/v1	OpenAI-compatible base URL
OPENCODE_MODEL	deepseek-v4-flash	Hosted model (any Zen model id)
PRIVATE_DOC_PATH	content/private-context.md	Path to the bio

Roadmap

See PLAN.md.

License

MIT