In-product assistants
An assistant inside your product that answers from each customer's own records and documents, under your brand.
RAG · Planning Sprint · 2 weeks · $5,000
RAG development. AI assistants that answer from your own data.
We build AI assistants, chatbots and search that answer from your product's own records and documents, cite where each answer came from, and only use what each user is allowed to see. Built for products with many customers, not a single demo.
Cited answers
Every answer linked to the passage it came from
Per customer
Each user searches only what they may see
Within a week
The Planning Sprint starts after you book
Answering from their own data at
Ask Ledgerly
Acme Dental workspace
What notice do we need to give to end the Lakeside lease early?
Searched only Acme Dental's documents this user can see
Illustrative example on a fictional product. Not client work.
What RAG fixes
No made-up answers. No one else's data.
Retrieval-augmented generation (RAG) lets an AI model answer from your own data: it first finds the right records and passages, then writes the answer from them. It fixes the two problems that stop most chatbots from shipping. Read how a RAG chatbot works.
Support assistants
Answers from your help center and product content, with a handoff to a person when the answer is not there.
Document search with citations
Questions answered across contracts, files and archives, with every answer linked to the passage it came from.
Multi-tenant RAG for SaaS
One assistant for all your customers, with each customer's data kept apart and each user's permissions applied to every answer.
Demo chatbot or production assistant
Scored. Cited. Permissioned.
Three things decide whether customers can rely on an assistant, and each one goes deeper in our guides. Below them, how a production assistant differs from a demo chatbot.
Tested on your own questions
Before release, the assistant is scored on real questions from your users, and failures are fixed before customers see it.
How to evaluate RAGCitations on every answer
Users see where each answer came from, so they can check it, and the assistant says so when the answer is not in the data.
How a RAG chatbot worksPermissions per customer and per user
Retrieval only searches what the user is allowed to see, so one customer's data never reaches another.
RAG securityA demo chatbot
It sounds right in a demo.
- Where answers come fromWhat the model already knows, or a few uploaded files
- How users can check itThey cannot
- When the answer is not thereIt guesses
- Whose data it searchesOne index for everyone
- How it is checkedA handful of questions in a demo
- Who answers for itYour team, once it is switched on
- After releaseYour team finds the problems
Your users cannot tell when it guesses.
A Datics production assistant
Every answer can be checked.
- Where answers come fromYour product's records and documents, found for each question
- How users can check itA citation on every answer, linked to the passage
- When the answer is not thereIt says so
- Whose data it searchesEach customer kept apart, each user's permissions on every search
- How it is checkedScored on real questions from your users before release
- Who answers for itWe own the delivery, to the done criteria agreed up front
- After releaseAnswer quality and cost monitored, and a two-week warranty
- Cited
- Per customer
- Scored on your questions
Try it
Ask it yourself.
Pick a question and who is asking. One answer is cited, one is not in the data, one asks for another customer's data, and one depends on an amendment.
1 · Pick a question
2 · Ask as
The same assistant, the same documents. What changes is what the question asks for and who is asking, and the assistant follows both.
Ask Ledgerly
Acme Dental workspace
What notice do we need to give to end the Lakeside lease early?
Cited answer
Ninety days' written notice, and only after the second year of the term [1]. An early end also carries a fee of two months' rent [2].
- [1] Lakeside lease.pdf · section 14.2
- [2] Lakeside lease.pdf · section 14.4
Searched 214 Acme Dental documents an office manager can see
Illustrative example on a fictional product. Not client work.
Want to click through a real product with AI built in? Try our working demo.
How it runs
Scored on your questions. Before your customers ask them.
Four steps, from a free call to an assistant your customers use. Every one ends with something you keep.
30 minutes
No chargeProduct Review
You show us the product and the questions your users ask. We tell you plainly whether RAG fits your data.
You keep
A straight answer on fit
Weeks 1 and 2
$5,000Planning Sprint
Starts within a week of booking. We set the data, the permissions, the questions it is scored on and the cost per question.
You keep
An assistant brief and a fixed quote
Every two weeks
Fixed quoteBuild Sprints
Built in your code and cloud, scored against real questions from your users, with a working demo every sprint.
You keep
A demo and the scores every sprint
When live
2-week warrantyIn production
Released with monitoring of answer quality and cost per question, and improved as your data grows.
You keep
The pipeline, question set and runbook
Sample evaluation run
Before release
Real questions from your users, each with the answer your team expects.
Answer matches the expected one46 of 50
Every claim cited to a passage49 of 50
Says so when the answer is not there8 of 8
Never shows another customer's data20 of 20
4 to fix before release
Amended leases: the assistant read the original clause, not the amendment.
Illustrative example on a fictional product. Not client work.
Want our engineer in your team?
Forward deployed engineers.
Our engineer joins your standups, repositories and reviews, and stays until the outcome is live. It starts with a one-week, $2,500 Outcome Sprint.
See forward deployed engineersAlready built a demo?
Start with the AI Scorecard.
It tests what you built on 50 of your own cases and ranks what to fix before it reaches every customer. $5,000, two weeks, and the fee counts toward production.
Take a demo to productionTry it
Not sure which start fits?
Four questions, and we show the start that fits your product, with what it costs.
- 01Is your product live, with customers using it?
- 02Have you already built an AI feature or agent?
- 03Do you know the outcome you need?
- 04Do you have engineers who would work alongside ours?
Four questions
Find the right way to start.
Answer all four and we show the start that fits your product, with what it costs. Your answers go with your booking, so the call starts where you are.
In practice
How it ran on our clients' own data.
Six assistants and search systems answering from the client's own documents, in the client's own systems, and how the work ran.
Document management
NenoBrief
Cited answers inside a live document system, with the original files never copied out of it.
How it ran
- Models self-hosted on the client's own GPU host
- Retrieval filtered by each user's access before any ranking
- Every finding from its test rounds fixed and rechecked with it
Healthcare
Datascape
One RAG pipeline reads compensation terms from any physician contract, inside its own HIPAA cloud.
How it ran
- Onboarded into its Slack, GitHub and AWS under a signed BAA
- Every change shipped as a versioned release it could test
- Each release left code, docs and a walkthrough in its account
Financial services
Cosmic AI
One chat that answers from filings, news, social posts and market data, citing the source behind each answer.
How it ran
- Data pipelines, retrieval and chat owned as the product team
- Data not cleared for users stayed in QA
- Every change through a separate QA environment first
B2B software
WikiTeq
Answers from the firm's own documents in its existing chat, with the source shown for every answer.
How it ran
- Scoping calls set the boundary first: one retrieval service, one filter
- Nothing changed in the chat product it already ran
- Reviewed and revised before delivery
Archives
NenoDoc Lexicon
Teams ask in French or English and get answers cited to the page, with no outside AI service.
How it ran
- Everything runs on the client's own AWS GPU instances
- Tested with the client's own documents and real users
- Declines instead of inventing a name, as the client's tests confirmed
B2B software
Amove
A Slack assistant that answers product, pricing and cloud cost questions from the company's own content.
How it ran
- Stored FAQ answers returned as written, before any model call
- Scope written and revised three times before any code
- Tuned from the review sheet the client kept while testing in its own Slack
More assistants on clients' own data
- Coast TechnologyDealer site shoppers ask in plain words and get real units from the dealer's own inventory
- FLZRA sales assistant that answers from the seller's own documents and files, at any hour
- TrailBlaze InnovationsMeal planning where code computes every calorie from food data and the model only phrases the plan
- Summers & CompanyPublic data structured into the knowledge graph behind an AI platform
- A regulated investment marketplaceAn assistant that answers only from the deal in view, and keeps nothing
- SovrenAIA chat where every claim carries a tier, a confidence and its source
What clients say
AI that answers from their own data.
The founders and CTOs whose data our assistants and pipelines work on, in their own words.
Datascape · RAG pipeline live, hired again
“Umair is a proven professional who knows what it takes to get things done. His engineers have a deep well of knowledge they can tap for tackling complex development problems. I'm looking forward to continuing work together.”
Summers & Company · Knowledge graph behind an AI platform
“What stood out to me was consistency: the work was built, tested, and delivered without noise.”
5.0 average client rating
Questions
RAG and AI assistant development, answered.
Adding other AI to your product as well? See AI integration services.
How much does it cost to build a RAG assistant?
It depends on how much data the assistant draws on, how many customers share it and how permissions apply. You can start with the $5,000, two-week Planning Sprint, which ends with a fixed quote for the build. The sprint fee counts toward it.
How accurate are RAG answers?
As accurate as the data and the evaluation behind them. We score the assistant on real questions from your users before release, show citations on every answer, and have it say so when the answer is not in the data rather than guess.
How soon can you start on a new AI assistant?
The Planning Sprint starts within a week of booking.
What if building the AI assistant takes longer than planned?
Within the agreed scope, that is on us: the fixed quote does not change. If you add to the scope, we agree the change and its price with you before any work on it starts.
What happens after the AI assistant is released?
Every assistant comes with a two-week warranty after release: anything that does not meet the agreed done criteria is fixed at no extra cost. Further work is quoted on its own, only if you want it.
Who is accountable for delivering the AI assistant?
We are. Datics owns the delivery of each assistant, with our own team and process, to the done criteria agreed in the Planning Sprint and at the fixed quote. You review the work and the demos; you do not manage the people doing it.
Who builds the AI assistant?
Senior Datics engineers who have shipped AI in production inside live products, backed by the wider Datics team on architecture and applied AI. If you would rather have our engineer work inside your own team, that is a forward deployed engagement.
What do your RAG development services include?
In-product assistants, support assistants, document search with citations and multi-tenant RAG for SaaS. Each one starts with a Planning Sprint that sets the data, permissions, test questions and cost per question, then is built in two-week sprints, scored on your own questions and released with monitoring.
Do you build AI chatbots?
Yes, when the chatbot answers from your own data. An AI chatbot built on RAG finds the right records and passages first, answers from them with a citation, and says so when the answer is not there, instead of guessing.
Are you an LLM development company?
We build LLM applications inside products, not foundation models. We work with models from AWS Bedrock, Azure AI and Google Vertex AI, and with self-hosted models on a client's own GPU host, as for NenoBrief and NenoDoc Lexicon, and choose the model for each use by testing it on your own questions.
Which vector database do you use?
The one that fits your stack. That can be the database you already run, such as Postgres with pgvector, or a managed option in your cloud. Retrieval quality depends more on how the data is prepared and permissioned than on the database brand.
Is our data private?
Yes. The assistant can run inside your own cloud account, or on your own GPU host with self-hosted models, the model services we use do not train on your data, and each answer only draws on what the user is allowed to see.
Can one assistant serve all of our customers?
Yes. In a multi-tenant product, one assistant serves every customer while keeping each customer's data apart and applying each user's permissions to every search. Getting this right is the main difference between a demo and a product.
Want your users to get answers from your own data?
Show us the product and the questions your users ask. In 30 minutes you will know whether RAG fits and what it would take.