Your assistant makes things up, and someone uploads files to it by hand every week.
An assistant over your own documents that answers only from approved sources, cites the passage behind every answer, says "not in the documents" when it doesn’t know, and stays in sync with where your files already live.
You have this problem if…
- The assistant invents answers: deals that don't exist, policies you never wrote, numbers that aren't in the file.
- Someone exports and uploads documents by hand, so the assistant is always a week behind.
- You can't tell which document an answer came from, so nobody trusts it for anything that matters.
What you get
- Automatic sync from Google Drive, SharePoint, Notion, your CRM or a SQL database
- Answers drawn only from approved sources, each citing the passage it used
- A refusal when the answer isn't in the documents, instead of a guess
- Escalation to a named person for questions the documents can't answer
- A 50-question acceptance set built with you, scored for correct citations and correct refusals
How it is done
- 01
Write the acceptance set first
Fifty questions with the answers you expect, including questions the documents can't answer, which the assistant must refuse.
- 02
Connect sources
Sync from where your files already live, respecting the permissions already set on them.
- 03
Index for retrieval
Documents are split along their real structure (clauses, tables, headings), so the right passage can be found and quoted.
- 04
Answer with evidence
Every claim in an answer is tied to a passage. Claims without support are removed, and if nothing is left the assistant says so.
- 05
Score and fix
The acceptance set runs against the assistant, and failures are fixed until it passes.
- 06
Hand over
You get the acceptance set, a runbook and a recorded walkthrough. The assistant runs in your accounts.
What usually goes wrong, and how we handle it
A bigger model doesn't stop invention
Most invented answers come from retrieval missing the right passage, then the model filling the gap. The fix is better retrieval, required evidence, and permission to refuse.
Refusal has to be tested
An assistant that never says "I don’t know" will guess. The acceptance set includes unanswerable questions, and the assistant has to refuse them.
Numbers come from code, not prose
Totals and comparisons are calculated from the source tables, not generated by the model.
Why you can trust us with this
Our founder built retrieval-augmented question answering over 10-K and 10-Q filings and earnings calls, with every answer traceable to its source. Report review time fell 95%.
His 2026 preprint on evidence-bound factual repair in retrieval-augmented answers describes the method used to keep each sentence tied to its evidence.
What we don't do on this package
- No assistants over medical records or other health information (PHI).
- No answers from the open web, unless you ask for it.
- No training or fine-tuning a model on your documents.
Questions
Can staff use it from Slack, Teams, ChatGPT or Claude?
Yes. The assistant can sit behind a Slack or Teams bot, a web page, or a connector in ChatGPT or Claude. Citations and refusals live in the retrieval layer, so they survive a change of front end.
What about documents only some staff should see?
Sync respects the permissions on the source, or we index only the folders you approve.
Who does the work?
Suresh Rajendran, our founder, does the engineering on every engagement himself. Nothing is handed to a junior engineer or a subcontractor.
Where does our data go?
We work inside your accounts: your cloud, your repository, your model-vendor keys. Where a vendor offers zero data retention we turn it on, and your data is never used to train a model. Our access is removed and credentials are rotated at handover.
How do we pay?
The first step is paid up front. Builds are 50% to start and 50% on acceptance. We invoice from Destello AI Incorporated in US dollars, payable by ACH or card. If you go ahead with a build within 30 days of the first step, its fee is credited.
What if the result doesn't meet the bar?
Acceptance criteria are agreed in writing before work starts, and the second payment is due only when they're met.
Will you sign our NDA?
Yes, usually within one business day.
Start with the single-source starter.
$1,200, credited to the build if you go ahead within 30 days. Book a 20-minute call, and you'll have a written scope with acceptance criteria within one business day.
Or email hello@destello.us or call (628) 202-5322. We reply within one business day.