AI Engineer Portfolio & Personal AI Workspace

The problem

A portfolio that only lists work goes stale the moment it ships, and the daily record of what was actually built lives scattered across notes, chats and memory.

Why it needed solving

Two things worth solving at once: recruiters and clients need a grounded way to ask questions about the work, and the author needs a low-friction way to capture daily engineering work from a phone without opening a task app.

The approach

A public portfolio with a Gemini-backed assistant answering strictly from verified profile content, and a private dashboard whose tasks can be created by sending an ordinary message — an LLM extracts the structured task from free text.

Architecture

• Public site — Next.js App Router, statically rendered content pages driven by a typed content layer.
• Assistant — Gemini called server-side only, grounded in a curated portfolio knowledge base, with prompt-injection resistance and no access to private data.
• Private dashboard — credential auth, task management, work journal and analytics.
• Message ingestion — Telegram webhook → authorized-sender check → Gemini structured extraction → database.
• Private assistant — question answering over the task history, behind authentication.

Implementation

• Content layer holds verified profile facts in typed modules so the site and the assistant answer from the same source.
• All model calls run server-side; no API key ever reaches the browser.
• Public and private data paths are separated at the database and route level, so the public assistant cannot reach work records.

AI/ML components

Challenges & how they were handled

What comes next

• Voice agent with speech-to-text and text-to-speech.
• Weekly work summaries generated from the task history.