AI Product Builder · Senior Product Manager
I turn ambiguous user problems into working AI products — from discovery and rapid prototyping to LLM integration, security, and deployment.
📍 Porto, Portugal
A self-hosted Telegram bot that turns posts from selected channels into concise, scheduled AI digests.
Currently used by 34 users with 21 active digests.
- Supports multiple daily and weekly digests
- Collects posts from public Telegram channels
- Handles timezones, retries, rate limits, and overlapping runs
- Uses token-aware chunking and model fallback
- Tracks runs and usage in PostgreSQL
- Uses Telegram as the complete product interface — an unused web dashboard was deliberately removed
Built with: TypeScript, Node.js, Telegram Bot API, PostgreSQL, Prisma, and OpenAI
Open Morningfeed Bot on Telegram →
A multi-agent Codex research system for Product-focused employer evaluation. It uses an orchestrator → Collector → Analyzer architecture, role-based reasoning, structured context handoffs, adaptive evidence sampling, and deterministic Python validation.
Built after reviewing public employer-research and career-agent implementations and identifying gaps in Product-specific evidence handling, multi-source validation, context isolation, incremental research, and auditability.
Focus: Agent Skills · agent orchestration · subagents · reasoning routing · context engineering · Python · evidence systems
A native macOS utility combining a persistent countdown timer with world clocks and desktop widgets.
| App · light | App · dark |
|---|---|
 |
 |
| Widgets · light | Widgets · dark |
|---|---|
 |
 |
 |
 |
 |
 |
- Searches the full macOS IANA time-zone database, including cities and seasonal designations such as CET/CEST, EET/EEST, and WET/WEST
- Shows each clock's live difference from the Mac time zone and handles DST automatically
- Shares one deadline-based timer between the app and interactive WidgetKit widgets
- Includes small, medium, and large clock widgets, plus a small timer widget
- Supports 11 interface languages and system light/dark appearance
- Works fully offline and ships as a self-contained Apple Silicon DMG
Built with: Swift, SwiftUI, WidgetKit, App Intents, UserNotifications, and XCTest
A self-hosted AI assistant for tailoring resumes and cover letters to specific job vacancies.
- Compares a resume with a vacancy and explains strengths and gaps
- Rewrites content without intentionally inventing experience
- Shows before/after changes and ATS keywords for human review
- Generates downloadable DOCX documents
- Supports English, Spanish, Russian, Portuguese, and Belarusian
- Includes privacy controls, anonymous sessions, rate limiting, and SSRF protection
Built with: TypeScript, React, Express, PostgreSQL, Drizzle ORM, and OpenAI
A native macOS interface built on top of the open-source Demucs music source separation model.
- Drag-and-drop audio import
- Configurable separation modes and output formats
- Live progress and recovery states
- Apple Silicon acceleration
- Distributed as a standalone DMG with bundled dependencies
Built with: Python, PySide6, PyTorch, Demucs, and FFmpeg