16 projects: AI apps in the stores, websites, stores and team products. Each with what was done, a link and a client review.
Work & reviews
16 projects — 16 five-star reviews
01 / 15
scroll→
Superconsciousvoice clone → claude → tts
01
AI · WellnessiOS · Android · Web
AI sessions in your own voice
What was done
React Native app and Next.js web
Claude pipeline → voice cloning → TTS
Cold start −45%, crashes −28%
★★★★★
A fairly tech-heavy project, and a key task was explaining what the product does without overloading the interface with technical details. That was handled well. Especially liked that we discussed not just visuals but the user journey itself. The product became much clearer from the very first screens.
We worked on Dzeny together with the design team. The task was tricky — make an AI product about mental health modern, calm and clear at the same time. We got exactly what we wanted. What we liked most is that it wasn't just drawing screens — the user flows and product logic were properly worked through. Happy with the result.
We came for iPlanly's design when it was clear the product lacked a coherent visual concept. Loved how the core flow was worked out: the user uploads a floor plan, picks parameters and gets interior options. The interface became much clearer, and the product now looks like a full modern service rather than just an AI tool.
It was crucial for us to show the AI result as clearly as possible. The team got that and built the interface around the visual result. Special mention for the user flow — uploading a photo, picking a style and getting the result now feel very consistent. Noticeably better than the original version.
We ordered the design for a tattoo sketch generator. We wanted it visually interesting without turning it into an overloaded "creative" interface. That was handled well. Especially liked how sketch variants and the generation process were designed. The result is a product you actually want to use.
A fairly large project: we had to think through not only logo creation but the whole brand-kit workflow. Liked that all these flows came together into one clear system. There were many iterations, but every change was handled smoothly, without repeated explanations. We ended up with an interface we can show clients without it feeling raw.
An interesting project with a rather specific audience, so nailing the product's atmosphere mattered. The visual style turned out nice, and the chat and character setup flows became much clearer. Also loved the attention to detail — things like screen states and transitions really shape how the app feels.
Structure for medical services, insurance and account
Trustworthy without the clinical look
Product pages and lead forms
★★★★★
One of the most complex projects in terms of content and flows: medical services, insurance and a personal account all at once. The team really understood the domain and helped bring it all into a clear structure. We needed the service to look trustworthy without resembling a typical clunky medical portal. That was achieved.
We worked on a fireplace and stove catalog site. The main goal was to make a large range easy to browse while showing the products properly. Liked the attention given to the catalog, filters and product cards. After the update the site feels much nicer on both desktop and mobile. Good work, especially given the amount of content.
We needed a store that looks modern without the usual e-commerce clutter. We got a clean, neat design. Especially liked the work on the mobile version — one of our key requirements. The team responded well to feedback and made changes quickly.
We needed the site to show the level of the studio's work first and foremost, so the focus was on cases and visual presentation. We got exactly the site we wanted: minimal text, maximum attention on the work. Loved the composition and portfolio presentation. The site itself has become part of our pitch materials.
We worked on a subscription workout platform with lots of content, categories and states, so the structure had to be built first. The team proposed clear navigation and a clear way to present workouts. Visually it's very neat and true to the product. Special mention for the mobile flow — that mattered to us.
The project needed a lot of interface work across many user flows. Liked that they didn't just go for a "pretty gaming design" but worked out the product logic first. Thanks to that the interface is rich yet clear. There were many iterations along the way, and the team went through them calmly with us.
The main challenge was the sheer amount of financial data and operations. We needed the app to be clear to an ordinary user, not just a finance expert. It turned out clean and logical. The core flows are well worked out, and the screens are much easier to read than in the original version.
A big project: partner dashboards and mobile flows had to be designed at the same time. What we liked most was the systematic approach. It never felt like each screen was designed in isolation — a single logic and design system gradually took shape. That mattered because we plan to keep growing the product.
Proven AI tools — deployed and tuned to your process
Open-source projects already removing routine for thousands of teams: documents, finance, email, media, hiring, development. I install them on your server, connect them to your services and adapt them to the job. Authors are credited and links go to the source code.
Documents · archiveopen source · ★ 5.9k
Automatic document processing and smart archive search
Paperless-AI · by clusterzx · MIT
Problem
Scans of invoices, contracts, letters and receipts pour into the archive. Each one had to be named by hand, with sender, type, date and tags. Hundreds of documents took hours, and finding a paper a year later was nearly impossible.
Solution
The service spots new documents in Paperless-ngx and sends their content to a language model: OpenAI, Ollama, DeepSeek, Azure or any compatible one. The model sets the title, tags, type, sender and date. A built-in RAG chat answers questions about the whole archive in plain language: “When did I sign the lease?”, “How much was the last electricity bill?”.
upload → webhook or scheduled scan → model analysis → metadata saved → semantic search
Before → after
Manual tagging of every documenttagged right after upload, no human needed
Search by file namesemantic search and plain-language questions
Inconsistent tagsone tag system built on the existing tags
The archive keeps itself in order. People only upload documents and ask questions.
Node.jsDockerOpenAI / Ollama / AzureRAGPaperless-ngx API
Every statement brings dozens of transactions with cryptic merchant names. For the budget to mean anything, each has to be categorised by hand. People put it off, and the budget drifts away from reality.
Solution
The service connects to Actual Budget, syncs accounts on a schedule and passes uncategorised transactions to a language model. For unknown merchants it looks them up online and adds that to the prompt. If no category fits, the model suggests a new one. A preview mode lets you check the decisions before applying them.
account sync → uncategorised transactions → model prompt with context → category and tag → retry the skipped ones
Before → after
Sorting every transaction by handautomatic processing on a schedule
A rigid category listnew categories suggested when needed
“I’ll sort it out at the weekend”the budget is always up to date
Bookkeeping no longer needs regular manual work, and the model’s decisions can be reviewed and rolled back.
TypeScriptDockerOpenAI / Anthropic / Google / Groq / OllamaActual Budget API
Many videos have no subtitles, and some are out of sync. The options were hunting on third-party sites, fixing timings by hand or watching without them.
Solution
The service transcribes speech in video and audio locally, on your own hardware, with Whisper and saves ready .srt files. It runs by webhook when a new file appears in Plex, Jellyfin, Emby or Tautulli, or works as a Whisper provider for Bazarr. The media server refreshes metadata and subtitles are available right away.
new video → webhook → queue → language detection → transcription → subtitle file → library refresh
Before → after
Searching and syncing subtitles by handgenerated with no human involved
Reliance on third-party sitesfully local, data never leaves
Part of the library without subtitlessubtitles across the whole library
Every new file in the library gets subtitles automatically, with no cloud services.
Email eats hours: sorting messages, unsubscribing from newsletters, fending off cold emails, remembering who to reply to and who owes a reply, writing the same answers again and again.
Solution
The assistant sorts the inbox into categories and drafts replies in the owner’s style in advance. Rules are written in plain text: “newsletters to archive, label invoices and save attachments”. It tracks emails awaiting a reply, blocks cold emails, bulk-unsubscribes from newsletters and saves attachments to Google Drive or OneDrive. It can be controlled from Slack or Telegram.
new email → model checks it against the rules → action: label, archive, draft, forward, webhook → reply tracking
Before → after
Sorting email by handsorted the moment it arrives
Replies from scratchready drafts in the owner’s style
Lost emailsevery email awaiting a reply is tracked
Email work comes down to reviewing and sending ready replies.
TypeScriptNext.jsVercel AI SDKPostgreSQLRedisGmail / Outlook API
Links, articles, notes and images pile up in dozens of places. Nobody has time to file them by hand, so the right thing can’t be found later.
Solution
The app saves a link, note or image, downloads the page content and hands it to a language model that tags it automatically. Images go through OCR. Works with cloud models and local ones via Ollama. Self-hosted.
save → fetch content → AI tagging → full-text search
Before → after
Folders and tags by handtags applied automatically
“I saved it somewhere”one base with content search
Dead linksan archived copy of the page
Saving takes one click, and anything can be found at any time.
The résumé gets rewritten by hand for every job: guessing the keywords applicant tracking systems (ATS) look for and matching the wording to the requirements.
Solution
The tool compares the résumé with the job text, shows a match score, finds missing keywords and skills and suggests better wording. It builds résumés and cover letters as PDF and works with 100+ models, local ones included.
résumé and job post → keywords → match score → suggested edits
Before → after
Editing by eyea measurable match score
Guessing keywordsa list of what’s missing
One résumé for every joba version for each position
Tailoring a résumé takes minutes instead of long manual editing.
Developers make dozens of commits a day, and messages often read “fix” or “update”. Such a history is nearly impossible to follow later.
Solution
A CLI reads the code changes and writes a meaningful message in the chosen convention, such as Conventional Commits. It runs as a git hook or in CI via a GitHub Action. Supports Claude, GPT and local models via Ollama.