For companies in Czechia and the EU · end-to-end delivery

Answers from your SQL data

Give staff a way to ask questions in plain language and check the source of each answer. Each project covers the path from a prototype on your SQL data to integration and handover.

A sample question is enough for a first discussion.

Where RAG can help

Start with a recurring question or a search problem in an existing tool.

Internal search

Staff repeatedly ask colleagues for information already held in SQL. They need search that respects their access permissions.

Support assistant

An operator needs an answer based on company data with a verifiable source. The system helps prepare an answer while making limitations visible.

Improve an existing solution

A prototype exists, but retrieval or answers fall short. Identify the causes and take an agreed use case through to deployment.

What the project includes

An agreed use case from SQL data review to integration and operational handover.

Discuss your project
  • A defined task, solution boundaries and access permissions.
  • A prototype using an agreed SQL data sample.
  • Integration with the selected interface and language model.
  • Checks against reference questions and agreed acceptance criteria.
  • Documentation and a data update procedure.

Scope, estimate and milestones are agreed before work begins. Model usage, infrastructure and ongoing support are estimated separately.

For a fixed report or an exact value, a SQL query may be enough. RAG is useful when people ask varied questions and need an explanation grounded in the records.

How RAG works

RAG gives a language model relevant company records to use when drafting an answer.

Fictional order A delayed shipment

A customer asks about an order

An account manager opens the order and passes the customer’s question to the assistant.

Why did only part of my order arrive?

From words to meaning

The model turns the question into a vector, a set of numbers that represents its meaning.

Find a note with a similar meaning

The search compares the question vector with two order records in SQL. Stock replenishment explains the delay. The delivery address only tells us where the order is going.

Awaiting new stock for the second batch. No dispatch date set.

Draft a reply from the note

The model writes a reply that the account manager can check against the warehouse note.

We’re waiting for stock to send the rest of your order. We don’t have a shipping date yet.
Warehouse note

Development and consulting

RAG design and implementation

  • Define the use case: an internal chatbot, knowledge search or a support assistant.
  • Review the SQL database, prepare data and design retrieval with access permissions in mind.
  • Build a prototype, evaluate answers, integrate the system and document operations.

What you receive: A system in the agreed environment, test results and a data update plan. Acceptance criteria are agreed before deployment.

RAG consulting and audit

  • Review an existing solution or requirements for a new one.
  • Examine retrieval, answer evidence, security constraints and costs.
  • Compare architectures and prioritise improvements.

What you receive: Written recommendations on what to change, how to evaluate the effect and which limitations to consider before investing in development.

From question to implementation

  1. Brief

    Identify users, tasks and example questions. Agree on constraints and the intended outcome.

  2. Data audit

    Review the SQL schema, data quality, retrieval capabilities and permitted access.

  3. Prototype

    Build a focused scenario on an agreed sample. Check whether RAG fits the task.

  4. Testing

    Evaluate retrieval and answers on reference questions, source links, refusals, latency and query costs.

  5. Implementation

    Connect the solution to the SQL database and agreed interfaces, configure access and hand over documentation.

  6. Support

    Under a separate agreement, update the index, investigate errors and evaluate quality after changes.

Technology choices

The solution uses a SQL database as its sole data store. Design starts with your database capabilities, data structure and access permissions. Reusing existing SQL infrastructure can simplify implementation; the scope follows the audit.

SQL database

One store for data and metadata. The audit establishes whether the client’s existing SQL database can be used.

SQL retrieval

Queries, filters and indexes are chosen for the database engine and question types. The approach is tested on an agreed sample.

Application logic

.NET / ASP.NET Core and Python for integration and data preparation. A language model drafts answers using information retrieved from SQL; its connection and processing terms are agreed separately.

Evaluation

Reference questions, retrieved-data and source checks, error analysis, latency and cost monitoring.

Formats and pricing

End-to-end project

From brief to implementation within an agreed scope. Estimate and milestones follow the data audit.

Priced against the agreed scope

Discuss this option

Consultation

A one-off session or hourly review. Receive recommendations and agreed next steps.

Priced against the agreed scope

Discuss this option

Ongoing support

Data maintenance, error investigation and improvements. Scope and response times are agreed separately.

Priced against the agreed scope

Discuss this option

Questions before we start

How long does implementation take?

Timing depends on data volume and condition, integrations and access requirements. After the brief and audit, we agree separate milestones for the prototype, testing and implementation.

How will our SQL data be processed?

We agree the processing environment before any data transfer. Access, hosting, logging and provider terms are reviewed. An anonymised description is enough for the first discussion; this form does not accept confidential files.

How do you choose a language model?

We compare options using your questions: answer quality, languages, latency, cost and data processing terms. The choice follows evaluation, rather than a preferred provider name.

What determines the price?

Data preparation, the number of sources, integrations, hosting requirements and evaluation scope. Development, model usage, infrastructure and support are estimated separately.

How is answer quality evaluated?

We agree reference questions and expected sources. Tests cover retrieval completeness, answer grounding, citations and behaviour when evidence is missing. RAG can still make mistakes; quality criteria and acceptable limitations are agreed in advance.

Can it run on-premise?

This can be considered. We first review available hardware, model licences, network and operational requirements, then assess feasibility and cost.

Can we use our existing SQL database?

We first review the engine, schema, data quality, indexes and access constraints, then identify any changes needed for retrieval. A SQL query may be enough for structured questions; RAG should be justified by the task.

What happens after launch?

You receive documentation and an agreed procedure for updating data. A separate support agreement covers access, monitoring and response times.

What happens after you get in touch

  1. Clarify the task

    Who will use the system, which questions it should answer and where the data is held.

  2. Agree the project

    Define scope, constraints, milestones, acceptance criteria and cost.

  3. Start with validation

    Check the core use case on a sample before expanding and deploying it.

Discuss implementation for your company

Describe your task in a few sentences. We will clarify the users, SQL database and constraints, then agree scope, milestones and cost before development begins.

Write an email

Email your name or company and a short question. You can also copy the address into any email service.