03 · What we build

Answers from your documents.

Retrieval-augmented generation lets an AI look things up in your own files before it answers, and show you exactly where each answer came from.

What it is

A general AI model knows a lot about the world and nothing about your business. Retrieval fixes that. When someone asks a question, the system first finds the most relevant passages in your handbook, policies, contracts, tickets or product docs, then writes the answer from those passages and cites them.

The model is rarely the hard part. The hard part is content: duplicated files, out-of-date versions, scans that cannot be read, and permissions that decide who is allowed to see what. That is where most of our effort goes, and where most projects succeed or fail.

In practice

What teams are building on their own knowledge

An internal help desk

New hires and busy staff ask the handbook, SOPs and policies instead of interrupting a colleague, and get an answer with the page it came from.

Sales and proposal support

Reps pull accurate specs, pricing rules and past proposal language in seconds, and drafts start from your real wording.

Customer-facing answers

A support assistant grounded in your product documentation, with a clear path to a person when it does not know.

Compliance and policy lookup

Staff check what the policy actually says, with the source passage, before they act.

Contract and document search

Ask “which clients have a 60-day termination clause?” and get a list with references to verify.

Onboarding assistants

A patient guide to how things work here, available at any hour, that improves as your documents do.

What we build

What you get

  • A content audit: what exists, what is stale, what is duplicated, what cannot be read
  • An ingestion pipeline for PDFs, Docs, wikis, email, tickets and spreadsheets
  • Search tuned to your documents and the questions people really ask
  • Access controls, so people only get answers from what they are allowed to see
  • Citations on every answer and an honest “I don't know” when nothing supports one
  • A test set of real questions, plus freshness syncing as documents change
How it runs

A typical engagement

  1. Audit the contentFind what is usable, what needs cleanup and who owns each source.
  2. Connect and indexBring the documents in with their permissions intact.
  3. Test with real questionsScore answers against what your experts say is correct.
  4. Launch and tuneWatch what people ask, fix gaps, and keep the index fresh.

A good fit if

  • Your team wastes time hunting for information that exists somewhere
  • You have a body of documents that people trust and reference
  • Answers must be traceable to a source

Probably not a fit if

  • The documents contradict each other and no one can say which is right
  • You need answers on data that does not exist in writing yet
Questions

Common questions

Does my data train the AI?

We use providers and settings whose terms keep your documents out of model training, and we tell you which ones. For the most sensitive material we can use models hosted in your own cloud.

How accurate is it?

We measure it. We build a test set from real questions, score the answers with your experts, and re-run it whenever something changes. Every answer shows its sources so people can check.

What if our documents are a mess?

Most are. That is normal, and it is why we start with an audit. Often a modest cleanup of the twenty percent that people actually use gets you most of the value.

Want help with knowledge systems?

Tell us what you are trying to fix. We will give you an honest read on whether this is the right move.

Just curious? coffee@aiveny.com

You have a question, a hunch, or an idea that is still half-formed. Write it however it comes out. No pitch, no pressure, no form to fill in.

Ready to build? tacos@aiveny.com

There is a real project, a deadline, or a process eating your week. Tell us what is going on and we will dig in with you.

Not sure which inbox? Pick either. They land in the same place.