I build SaaS products and marketplaces, from idea to production.

Role: Co-founder, full-stack and AI

Cafes, bars and restaurants often track stock by hand, which leads to inaccurate counts and slow reconciliation. As a co-founder I built an AI-powered inventory SaaS that reads invoices and delivery notes, processes unstructured data, tracks recipe-based consumption and uses a unified PCS measurement model to keep counts consistent. Hospitality teams see what came in, what was used and what remains, without spreadsheets or recounting.

• TypeScript
• Supabase
• PostgreSQL
• AI document processing

Role: Back-end

Optilium is a command center for freight carriers: one place to track drivers, manage loads and match available capacity. I built the multi-tenant backend with real-time ELD driver tracking, load-board integration, a matching engine, scheduled jobs, load expiration and simulators for testing before the real ELD and load-board APIs were connected.

• NestJS
• Prisma
• Supabase
• PostgreSQL
• TypeScript

Role: AI automation and backend

A team managing projects across several tools had to hand off every change manually. I built an event-driven automation engine for a multi-tenant SaaS: user actions trigger rules that send emails, post to Slack, create Google Calendar events or call webhooks, with full status tracking (pending, processing, completed, failed). I also added an AI document analysis system (RAG). Every integration executes for real, with no mocked logic.

• Supabase
• PostgreSQL
• Slack
• Google Calendar API
• Resend
• RAG

Role: Full-stack

WALAR (World Accuracy Landing Ranking) needed to move its scoring off spreadsheets into a platform athletes, organizers and the public can use. I built a web application with a React/TypeScript frontend and a Supabase/PostgreSQL backend. The scoring engine (jump-distance lookup tables, placement bonus points, multipliers for national team members, juniors and medalists) is calculated server-side in PostgreSQL functions and feeds a rolling five-year leaderboard with filters. It includes database schema design, role-based RLS, GDPR-compliant consent for public athlete data, CSV/Excel import tooling and an admin panel.

• React
• TypeScript
• Supabase
• PostgreSQL

Role: Back-end

An aesthetic clinic needed to answer patient questions on WhatsApp instantly, without risking a wrong or hallucinated answer reaching a patient. I built a WhatsApp AI support agent: RAG-grounded answers from the clinic's real FAQ content, a mandatory human-in-the-loop approval layer (every AI draft must be approved, edited or rejected before sending) and AI tool-calling (MCP) so the model decides when to book an appointment. It includes webhook signature verification, idempotent message handling and full audit logging. No AI message reaches a person without a human decision.

• Supabase
• RAG
• MCP
• WhatsApp

Role: Back-end

A finance SaaS needed to turn raw income and expense data into insight users would act on, not another spreadsheet. I built the AI analysis layer that detects spending patterns, flags financial risks and generates personalized natural-language recommendations from structured data. It is deliberately not a chatbot but an automated system triggered by real user data. The product includes a financial data model (income, expenses, savings, debt), Stripe subscriptions and role-based gating that locks AI features behind a Pro tier.

• Supabase
• OpenAI API
• Stripe

Role: Full-stack

Users tracking daily habits (sleep, hydration, movement, mindfulness) had raw data but no way to turn it into insight without manual effort. I built an automated pipeline that computes real-time analytics from time-series activity logs and triggers AI-generated behavioral insights tailored to each user's patterns, regenerated whenever new data comes in. The OpenAI integration does structured analysis rather than open-ended chat.

• Supabase
• OpenAI API

Role: Full-stack

A business needed a fast, trustworthy way to get answers from its own documents instead of manually searching PDFs, plus a way to compare two documents for missing items or mismatched terms. I built an AI document intelligence tool with two features on a shared retrieval foundation: a chat that answers with citations back to the source document, and an AI-generated comparison that flags differences and produces a structured report. Every answer stays verifiable and traceable to its source.

• RAG
• Supabase
• API integration

A presentation website for a villa in Istria: multiple languages, a gallery of rooms and surroundings, amenities and location, FAQ, reviews and booking.