HR & policies
Leave, benefits and expenses, answered from the current policy.
Knowledge assistants · RAG
Every answer your team needs.
With the page it came from.
An assistant for your staff that answers from your policies, procedures, contracts and manuals, in Hebrew and English, shows the source, and respects who may see what.
Ultrabot, built by MLAIA Data ScienceLed by Dr. Yochai Edlitz · Ph.D., Weizmann Institute
Where it helps
The answer exists somewhere.
Finding it takes too long.
IT & internal helpdesk
How-to answers from your guides, before anyone opens a ticket.
Sales & customer service
Answers for your agents, in approved wording.
Legal & contracts
Find clauses and compare versions, with a lawyer deciding.
Operations & procedures
Step by step from the manuals, on a phone on the floor.
Onboarding
New staff ask instead of searching.
RAG in production
Retrieval is the product.
The model is the easy part.
Most RAG projects that fail, fail at search, not at the model. We measure retrieval first.
Hebrew-aware search
Keyword and semantic search combined, tested on Hebrew and mixed Hebrew-English documents.
Permissions
People get answers only from documents they may already open.
Sources on every answer
Each answer cites the document and passage. No source, no answer.
Measured
A test set of real questions, scored for finding the right passage and for the answer, rerun on every change.
Typical sources & stackSharePoint & OneDriveGoogle DriveConfluencePDFs & scans (OCR)Amazon Bedrock Knowledge BasesOpenSearch & pgvectorSlack & TeamsOn-prem open models
Data & hosting
Your documents stay
in your environment.
Deploy in your own AWS account with Amazon Bedrock, or fully on-premises with open models when data may not leave your network. See also sovereign AI and on-prem LLMs in Israel and small language models on-premises.
A practical delivery path
One team and
one document set first.
Pick the team & documents
One team, its key documents, and an owner for the content.
Build a test set
Real questions with known answers.
Pilot
Measure retrieval and answers; fix the content gaps it reveals.
Roll out
More teams and sources, with monitoring.
Before we start
Good questions. Straight answers.
What is RAG?
Retrieval-augmented generation: the assistant first searches your documents, then writes an answer from the passages it found and cites them. It answers from your content rather than from the model's general knowledge.
Does it work with Hebrew documents and scans?
Yes. Hebrew and mixed-language documents are indexed with Hebrew-aware search, and scanned documents go through OCR first. We test retrieval on your own documents before rollout.
Can it respect our permissions?
Yes. Each answer is built only from documents the user is allowed to see, using your existing access groups.
Can it run without sending data outside?
Yes. It can run in your own AWS account, or on-premises with open models, so documents and questions stay in your environment.
How do you know it works?
We build a test set of real questions with known answers and score every version on it, both for finding the right passage and for answering correctly.
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.