Internal search
Staff repeatedly ask colleagues for information already held in SQL. They need search that respects their access permissions.
For companies in Czechia and the EU · end-to-end delivery
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.
Start with a recurring question or a search problem in an existing tool.
Staff repeatedly ask colleagues for information already held in SQL. They need search that respects their access permissions.
An operator needs an answer based on company data with a verifiable source. The system helps prepare an answer while making limitations visible.
A prototype exists, but retrieval or answers fall short. Identify the causes and take an agreed use case through to deployment.
An agreed use case from SQL data review to integration and operational handover.
Discuss your projectScope, 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.
RAG gives a language model relevant company records to use when drafting an answer.
An account manager opens the order and passes the customer’s question to the assistant.
Why did only part of my order arrive?
The model turns the question into a vector, a set of numbers that represents its 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.
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
What you receive: A system in the agreed environment, test results and a data update plan. Acceptance criteria are agreed before deployment.
What you receive: Written recommendations on what to change, how to evaluate the effect and which limitations to consider before investing in development.
Identify users, tasks and example questions. Agree on constraints and the intended outcome.
Review the SQL schema, data quality, retrieval capabilities and permitted access.
Build a focused scenario on an agreed sample. Check whether RAG fits the task.
Evaluate retrieval and answers on reference questions, source links, refusals, latency and query costs.
Connect the solution to the SQL database and agreed interfaces, configure access and hand over documentation.
Under a separate agreement, update the index, investigate errors and evaluate quality after changes.
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.
One store for data and metadata. The audit establishes whether the client’s existing SQL database can be used.
Queries, filters and indexes are chosen for the database engine and question types. The approach is tested on an agreed sample.
.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.
Reference questions, retrieved-data and source checks, error analysis, latency and cost monitoring.
From brief to implementation within an agreed scope. Estimate and milestones follow the data audit.
Priced against the agreed scope
Discuss this optionA one-off session or hourly review. Receive recommendations and agreed next steps.
Priced against the agreed scope
Discuss this optionData maintenance, error investigation and improvements. Scope and response times are agreed separately.
Priced against the agreed scope
Discuss this optionTiming 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.
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.
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.
Data preparation, the number of sources, integrations, hosting requirements and evaluation scope. Development, model usage, infrastructure and support are estimated separately.
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.
This can be considered. We first review available hardware, model licences, network and operational requirements, then assess feasibility and cost.
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.
You receive documentation and an agreed procedure for updating data. A separate support agreement covers access, monitoring and response times.
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.
Email your name or company and a short question. You can also copy the address into any email service.