Skip to content
Sun Software logo with the tagline Engineering Beyond Code
RAG & Knowledge-Based AI

RAG development that turns your business knowledge into grounded AI answers.

Our RAG development services build retrieval-based AI that helps authorized users find information in relevant, maintained sources, instead of relying only on a model’s general knowledge.

service-briefScoped proposal
  • CategoryAI Agents & Automation
  • EngagementScoped proposal
  • First stepDiscovery call
  • Reply withinOne business day
01

50+

Countries our clients come from

Mostly the US, UK, Germany, Canada and Australia.

02

1,685

Reviews on Fiverr

Public, and open for anyone to read.

03

16

People on the team

Working together since 2020.

04

4.9

Out of 5 on Fiverr

Average across all reviews.

The answer is in your files. Can your AI find the right one?

An assistant connected to your documents can still be wrong. Research shows how often.

  • 62%

    of workers say they spend too much time searching for information

    What we doWe build one place to ask, connected to the documents and systems where your answers live.

    Microsoft, 2023
  • 17%+

    of answers from specialist legal AI tools were wrong or made up, even with sources

    What we doWe test the assistant on real questions from your team, fix what it gets wrong and test again after every change.

    Stanford HAI, 2024
  • up to 24%

    of summaries contained facts the document never stated, depending on the AI model

    What we doWe show the passage behind every answer, and the assistant says so when it cannot find one.

    Vectara leaderboard, 2026
  • 16%

    of business-critical data is shared with more people than it should be

    What we doWe make the assistant follow your existing access rules, and test it with different users.

    Concentric AI

What a custom RAG system needs to be trusted.

Five problems businesses meet with AI over their own documents, and what we do about each.

  1. Problem 01 · Searching

    Staff lose time hunting for answers that already exist somewhere in your files.

    What we do

    We build one place to ask, connected to the documents and systems where the answers live.

    • One place to ask
    • Connected to your sources
    • Answers in plain language
  2. Problem 02 · Sources

    The assistant gives a confident answer, or cites a document that does not say what it claims.

    What we do

    Every answer shows the passage it came from, and the assistant says so when it cannot find one.

    • The source shown with each answer
    • Says when it does not know
    • No answer without a source
  3. Problem 03 · Old documents

    Old, duplicate and conflicting files mean the assistant answers from the wrong version.

    What we do

    We choose with you which documents count, remove duplicates, and keep the assistant updated as files change.

    • Agreed list of sources
    • Duplicates removed
    • Kept up to date
  4. Problem 04 · Access

    The assistant shows people information they should not see.

    What we do

    The assistant follows your existing access rules, so each person gets answers only from what they may open.

    • Your access rules respected
    • Tested with different users
    • Sensitive files kept out
  5. Problem 05 · Testing

    The assistant goes live without being tested, and you are responsible for what it says.

    What we do

    We test it against real questions from your team, fix what it gets wrong, and test again after every change.

    • A set of real test questions
    • Results you can read
    • Re-tested after changes

Real assistants. Real wrong answers.

Two public cases show why sources and access rules matter. Neither is our client.

  • Court sanction · June 2023

    Lawyers filed court cases that did not exist

    What happened

    In Mata v. Avianca, lawyers filed a brief written with ChatGPT that cited court decisions that were never made. A New York judge sanctioned them with a $5,000 penalty.

    What we do about it

    Our assistants answer only from your real documents and show the passage, so every answer can be checked.

    Source: CNBC
  • Vendor bug · early 2026

    An assistant summarised emails marked confidential

    What happened

    A code error made Microsoft 365 Copilot Chat summarise emails that carried confidentiality labels. Microsoft fixed it and said nobody saw information they were not already allowed to see.

    What we do about it

    We test access rules at the search step with different users, and test again after every change.

    Source: BleepingComputer

Our own work

Live products we built. Open them and judge the work for yourself. Work for private clients is shared on a call, with the client's agreement.

See all case studies

Answers you can check.

What every knowledge assistant we build comes with.

  • 01

    A source with every answer

    The passage it came from, one click away.

  • 02

    Your access rules

    People see only what they may already open.

  • 03

    Kept up to date

    New and changed files are picked up.

  • 04

    A test set

    Real questions from your team, with results.

  • 05

    An honest “I don't know”

    No answer when no source supports one.

  • 06

    A record of questions

    See what people ask and where it falls short.

Why choose Sun Software as your RAG development company?

The model is the easy part. The documents are the work.

We start with your documents, not with the model. We agree which sources count, show the source with every answer, follow your access rules, and test the assistant on real questions before your team relies on it.

The Sun Software team

  • 01

    We sort the documents first

    Old and duplicate files are dealt with before the assistant reads anything.

  • 02

    No source, no answer

    The assistant shows where each answer came from, or says it does not know.

  • 03

    Tested with your real questions

    You see the results before it goes live, and again after each change.

In their words, not ours.

No public review names this service yet. These are recent reviews about how we work, word for word from Fiverr.

  • Great....very professional, delivered on time and exactly as I needed. Exceptional. I always find Sunsoft very easy to work with, always helpful and I'll continue using their services for the foreseeable future. No need to go anywhere else.
    Client in United KingdomOctober 2026 · Fiverr review
  • Excellent experience from start to finish. Communication was great, responses were always prompt, and the work was completed quickly. They captured the exact look and feel I wanted on the first try and delivered excellent value for the money. I would gladly work with them again and highly recommend their services.
    Client in GermanySeptember 2026 · Fiverr review
  • Choosing SunSoft was exactly the right decision. Their professionalism, extensive experience, and attention to detail have really impressed me so far. It was a complex project that exceeded my expectations! I look forward to continuing to work with SunSoft!
    Client in CzechiaAugust 2026 · Fiverr review

Questions about knowledge-based AI

What is RAG, and what do RAG development services include?

RAG stands for retrieval-augmented generation: the AI looks up passages in your own documents before it answers. We build that as a knowledge assistant where every answer shows its source, your access rules are respected, new and changed files are picked up and a test set of real questions shows how well it works.

Can the system guarantee every answer is correct?

No. We reduce errors with citations, evaluation and the option to decline to answer, and keep human verification where it matters.

Can it respect who is allowed to see what?

Yes. Access rules are part of the design, and we test that users only retrieve what they are permitted to see.

What decides the price of knowledge-based AI?

Mainly content, access rules and accuracy needed. We agree the scope and the price in writing before work starts.

What decides the cost of a RAG system?

Most of the cost is in preparing the content and proving that the answers are right. Three things decide it. Content: How many documents, in which formats and in what condition. Access rules: Whether different users may see different sources. Accuracy needed: How much testing the answers need before people rely on them.

What happens after I contact you?

We reply within one business day. Then the work runs in 6 steps: Choose sources, Prepare content, Build retrieval, Build answers, Evaluate, Maintain. Each step ends with something you can review.

  1. 1Discuss a Knowledge AI System
  2. 2We reply within one business day
  3. 3Discovery call
  4. 4Step one: Choose sources

Ready to talk about knowledge-based AI?

We begin with your knowledge sources and the decisions your users need to make.

  • The first conversation is free
  • NDA before you share anything
  • A reply within one business day