// AI transformation
Expander
AI and digital tools where they save time: documents, voice, training and simple apps.
Client projectExpander
Expander is a financial advisory firm. Its advisors' work revolves around documents, client conversations and ongoing training. We build AI and digital solutions for the company: each one solves a specific problem, and together they change how everyday work gets done.
// What we delivered
- →Reading documents from a phone photo: straightening, OCR and data in the form
- →A mobile app you can operate by voice
- →Speech recognition and language models on our own GPUs
- →Training recordings generated from text with speech synthesis
- →Simple apps: merging reports and matching bank transfers to accounting
// how it works
01 / 04 · Document reading
Documents that read themselves
A phone photo of a document is all it takes. The system finds the page edges, corrects the perspective, reads the text and moves the right values into the form fields. A person reviews the result instead of retyping it. Reading a note takes 12 s instead of over 8 min.
02 / 04 · Voice control
An app you operate by voice
Instead of tapping through small fields on a screen, you just say what needs to be saved. Speech becomes text, text becomes structured data, and before saving, the app shows what it understood. Speech recognition runs on our own GPUs.
03 / 04 · Speech synthesis
Training narrated straight from text
A lesson starts as a script. The text is normalized first, so numbers and abbreviations sound natural, then speech synthesis turns it into narration and the recordings come together as a training module. Fixing the material means editing text, not booking another studio session. We are developing our own speech synthesis model.
04 / 04 · Simple apps
Simple apps that take work off people's plates
Not every problem needs AI. Reports from several sources, each in a different format, merge into a single summary. Bank transfers match themselves to accounting entries, even with a typo in the name or swapped digits in the account number. Unclear cases go to a person.
// Facts
- Time to read a note with OCR (previously over 8 min)
- 12 s
Technologies used
Next.js
React Native
Expo
Payload
MongoDB
TypeScript
Computer vision
LLM serving
Local LLMs
Speech-to-text
Kubernetes