Guided onboarding
Explains features in context, when the user needs them.
In-app assistants
An assistant inside your product.
Not a chat bolted on the side.
We build assistants into web and mobile apps. They answer from your product data, take actions through your own API with the user's permission, and can talk, in Hebrew and English.
Ultrabot, built by MLAIA Data ScienceLed by Dr. Yochai Edlitz · Ph.D., Weizmann Institute
What it does
Users ask.
The app does it.
The assistant works inside your screens, with the same data and permissions the user already has.
Actions through your API
Search, create and update, limited to the user's permissions, with confirmation for anything that changes data.
Answers from your data
Help docs plus the user's own records.
Voice in the app
Talk instead of typing, on mobile or when hands are busy. See voice agents.
A face for your product
A talking avatar for the assistant. See AI avatars.
Multi-tenant SaaS
Each customer's data and settings kept apart.
Engineering
Built like the rest
of your product.
An assistant that acts for users needs the same care as any other feature that writes data.
Typed tools, not screen-scraping
The assistant calls tools mapped to your existing endpoints.
Guardrails
Limits on what it may do, confirmations, and an audit log of every action.
Latency & cost
Model choice, caching and streaming tuned per feature; cost per active user tracked.
Evaluation
Test conversations rerun on every release.
Typical stackClaude, GPT & GeminiAmazon BedrockLangGraphReact & React NativeiOS & AndroidREST & GraphQL APIsStreaming speech (ElevenLabs, Amazon Transcribe)
What we are building
A voice assistant with a face, inside a healthcare app.
Our current work includes a Hebrew and English voice assistant with a lip-synced avatar inside a healthcare app, currently in development, running Claude on Amazon Bedrock with Amazon Transcribe.
We describe client projects in general terms. Details are shared in conversation, subject to confidentiality.
A practical delivery path
One job first.
Then the next feature.
Pick the jobs
Which user tasks the assistant takes on, and which actions it may call.
Prototype in your app
A working version against your staging API.
Beta
A user group, with feedback and measurement.
Ship & operate
Monitoring of quality, actions, latency and cost.
Before we start
Good questions. Straight answers.
Do we need to change our backend?
Usually very little. The assistant calls your existing API through a thin tool layer; we add endpoints only where an action isn't exposed yet.
Which models do you use?
We choose per feature: Claude, GPT or Gemini, through Amazon Bedrock or directly, or open models when the data or the cost calls for it. The model can be changed later without rebuilding the app.
Can it work on mobile?
Yes, in iOS and Android apps and on the web, by text or voice. Avatars can render in the app or stream as video.
Who owns the code and the data?
You do. It runs in your cloud account, and conversations and logs stay under your policies.
Keep exploring
Related reading & tools.
Talk to a person
Tell us what
you’re working on.
Share the problem, the data you have and what success would look like. We’ll discuss a practical next step.
Please don’t include confidential datasets, credentials or patient information.