drafta
SaaS · Next.js, Supabase, Mistral, OpenAI
SaaS for businesses: AI drafts replies to customers across messengers and social networks, you just press Send.
drafta is a SaaS app that makes customer communication easier for businesses. It takes over the job of an agent who runs the company’s social networks and messengers: it drafts replies and publishes content.
Challenge
Customers message small businesses on Telegram, WhatsApp, Instagram, Facebook, by email and in the website chat. The owner keeps switching between apps and answers every message by hand, leaving no time for social media.
Solution
All channels connect to a single inbox. For every request, AI (Mistral, OpenAI) drafts a reply using the business context: calendar, spreadsheets and knowledge base. The same app lets you create posts and publish them straight to social networks. Stack: Next.js and Supabase.
Result
The user only reviews the draft and presses one button to send it. The final decision always stays with a human, and conversations and posts across all channels are handled in one place.
Mira
AI automation · n8n, Next.js, Supabase, HubSpot, Telegram, OpenAI, Mistral
An AI travel search assistant: tell the chat where you want to go and on what budget, and agents find matching flights and hotels.
Mira is an AI chatbot for trip planning. The user describes in their own words where they want to go, how many people are travelling, the budget and any wishes, and AI agents search for matching flights and hotels.
Challenge
Planning a trip means hours of comparing flights and hotels across different websites, typing the same parameters into filters again and again and keeping all the options in your head.
Solution
The logic runs on n8n: an orchestrator agent collects the trip parameters, asks for missing details and launches separate flight and hotel search agents. Models: OpenAI and Mistral. Answer quality is checked with automated evaluations on a test dataset. The chat UI is built with Next.js, data lives in Supabase, with HubSpot and Telegram integrations.
Result
A conversation instead of filters: the user gets up to five matching flight and hotel options with price, rating, conditions and a short explanation of why each one was chosen.
Spirit FairPlate Shop
Web · Next.js, TypeScript, PostgreSQL, Self-hosted
A portal for Spirit GmbH: customers upload a packaging design and get the area per colour and the flexo plate price in a few clicks.
Spirit FairPlate Shop is an online portal I built for Spirit GmbH, a maker of flexographic printing plates for corrugated board. The customer uploads a packaging design, and the system splits it into colours, measures the area of each colour and calculates the order price.
Challenge
The price of a flexo plate depends on the plate area needed for each colour of the design. Spirit’s customers, packaging manufacturers, need that price fast to quote their own clients, while breaking down a design and measuring areas by hand takes time.
Solution
A portal built with Next.js and TypeScript, with a PostgreSQL database on a self-hosted server. An uploaded design is automatically separated into colours, the area of each colour is calculated, and the price is derived from it. Calculations can be saved and ordered once the client confirms, and without registering you can upload a file or use sample data to get a demo calculation by email.
Result
A few clicks instead of a manual request: the customer uploads a product template and immediately gets a detailed breakdown of technical areas per colour and the order price.
Benefita
Web · Next.js, Supabase, PWA
A cooperative loyalty programme for local businesses: points earned at one venue are spent on discounts at others.
Benefita is a platform that brings local businesses together in a shared loyalty programme. Users earn points by completing tasks at some venues and spend them on discounts and gifts at others, while businesses gain new customers from their network partners without an advertising budget.
Challenge
Small businesses struggle to attract new customers: advertising is expensive, and their own loyalty programme only keeps those who already came. A café, salon or workshop needs a way to get customers from nearby businesses and motivate them to come back.
Solution
Users find venues in the app’s showcase and complete tasks for points. The simplest is to visit a venue and scan the NFC chip at the counter to check in. Other tasks include posting about the venue on social media, following an account, inviting a friend and turning on notifications. Points are exchanged for discounts and gifts offered by participating venues. Stack: Next.js, Supabase; the app runs as a PWA, with no app store install.
Result
Customers of one venue discover others and come for their discounts, and every visit and social media post brings the business a new audience. For users it is a single bonus programme across the city instead of a dozen separate cards.
https://benefita.de/Product Card AI
AI automation · Gemini, React, API, Vercel
An AI assistant for sellers: turns an amateur product photo into a ready listing for Amazon, Etsy, eBay or Shopify.
A chatbot that turns a raw product photo into a ready marketplace listing. Just send a snapshot, say a slightly creased dress lying on a bed, and the assistant generates professional photos, a product title and a description.
Challenge
To list a product on a marketplace, a seller needs good photos, a compelling title and a description with key features. A photo shoot with a model and copy tailored to every platform cost time and money, especially with many products.
Solution
The user uploads a photo to the chat, and the assistant recognises the product, generates a set of images (on a clean background, in an interior and on a model “wearing” it) and writes a title, a description and a list of key features. Buttons choose the target platform (Amazon, Etsy, eBay, Shopify and others) and the language: English, German or Russian. Stack: React, Gemini models via API, hosted on Vercel.
Result
One amateur snapshot becomes a ready product listing in minutes: a set of photos, a title and a description that can be downloaded or edited before publishing.
Gastronomie AI
AI automation · LiveKit, ElevenLabs, Next.js, PostgreSQL, Self-hosted
A voice assistant for a restaurant: answers calls, books tables and takes takeaway orders, handing complex questions to a staff member.
An AI voice assistant and portal for a restaurant. A guest calls the restaurant’s regular number, and the assistant answers questions, books a table or takes a takeaway order. Everything it does goes straight into the restaurant portal, which is connected both to the assistant and to the website.
Challenge
At peak times restaurant staff can’t keep up with calls: guests can’t get through to book a table or order food, and the restaurant loses orders. Bookings and orders also come from different channels (phone, website and on site) and have to be combined by hand.
Solution
The voice assistant takes calls on the restaurant’s number: it answers guests’ questions, books tables and takes takeaway orders. If it can’t perform an action or answer a question, the call is transferred to a staff member. The second part of the project is the restaurant portal: a guest website with the menu, table booking and takeaway ordering, and a staff dashboard with a floor plan and real-time bookings and orders from all channels. Stack: LiveKit and ElevenLabs for voice, Next.js and PostgreSQL on a self-hosted server for the portal.
Result
The restaurant answers every call, even when the staff are busy. Bookings and orders from the voice assistant, the website and the dining room come together in one place, and each entry shows which channel it came from.
Lingvy
Software · Telegram Mini App, React, Supabase, Vercel
A German learning app built on books and films: reading with translation, audio retellings, a personal vocabulary list and word exercises.
Lingvy is a Telegram Mini App for learning a foreign language through texts based on well-known books and films. The user reads and listens to stories, translates unfamiliar words, collects them in a personal vocabulary list and practises them in exercises.
Challenge
Textbooks and flashcard apps get boring fast: words are memorised out of context and forgotten just as quickly. The idea was to learn a language through engaging stories while training reading, listening and vocabulary at the same time.
Solution
The app has a library of texts based on books and films, split into chapters. While reading, you can highlight a word or phrase, see its translation and add the word to your vocabulary list with one tap. Texts can be listened to as audio retellings, not just read, and the video section offers clips with German subtitles. Words from the list are practised in exercises: fill in the blank, listen and spell the word. Stack: a Telegram Mini App built with React, data in Supabase, hosted on Vercel.
Result
The user learns the language through stories they enjoy rather than isolated words. One app trains reading, listening and word retention, and the vocabulary grows right while reading.
