Applied AI by Tecela

When a bet needs software,
Tecela builds it.

Custom applications, AI agents and the data work underneath, built inside your systems with your team and delivered with source code and documentation. Our default advice is to buy what already exists, so we build what you cannot buy.

Request a 30-minute call Not sure what to build? Start with the audit

01 · Custom software

When the process
outgrows its tools.

Applications that bring rules, data and people into one workflow. AI joins where a concrete task needs it.

When it fits

  • A spreadsheet runs the operation, with no record of who changed what.
  • Customers and patients ask by message because there is nowhere to look it up.
  • Everyone sees everyone's data, with no split by user, role or company.
  • The off-the-shelf system covers 80% of the process and stalls on the rule that is yours.

What we have built

  • A financial portal for a medical practice, on mobile, using the same calculation base as the team's own dashboard.
  • An installable app with checkout, an orders back end and an admin panel.
  • A multi-account finance app with per-account isolation and a reviewed statement import.
  • An installable app that works offline and keeps its data on the device.

How we check it holds

  • One calculation base. The same rules feed the client view and the team view.
  • Permissions by user and organization, enforced at the data layer, not hidden behind a button.
  • Automated tests on the rules that matter, so a change cannot break one quietly.
  • Production with its own database, authentication and domain.

02 · AI agents

Repeated work needs
context and control.

Agents that consult sources, organize information and support actions inside the process. Limits, code checks and human review are set by the risk of the task.

When it fits

  • The answer exists, spread across meetings, documents and systems.
  • The team copies, summarizes and checks the same things every week.
  • People decide without the history of what already happened.
  • You tried AI and cannot check where an answer came from.

What we have built

  • A document reader that imports invoices and files, applies the calculation rules and leaves the review to the team.
  • A proposal writer that drafts from the conversation, with a preview and a review step before anything is sent.
  • A market researcher that reads video and comments and returns language, objections and themes, with sources.

How we check it holds

  • Numbers come from code. The model reads and interprets; it does not invent figures.
  • Every suggestion points to the document, record or passage behind it.
  • People approve, adjust or discard at the points that matter, before anything goes out.
  • Models are chosen by test on your own cases, for quality, cost and speed. Demos do not get a vote, and swapping a model is configuration.
  • Every run is logged: inputs, tools used and result.

03 · Data and integrations

Scattered data
slows every decision.

We connect sources, organize permissions and deliver data where the team works. The same base lets AI work with context and limits.

When it fits

  • CRM, ERP and spreadsheets each hold a piece of the same information.
  • Someone exports, pastes and checks the same numbers every week.
  • Anyone can see everything, with no split by company, role or user.
  • A dashboard shows a number and nobody knows where it came from.

What we have built

  • Integrations with corporate CRM and ERP systems through REST, OData and webhooks.
  • From invoice files to monthly close: imports in XML, ZIP and spreadsheet formats from different issuers, with a calculation engine separate from the interface.
  • A store connected to CRM and ERP, where orders and leads arrive automatically, with cart recovery and email sequences.
  • Searches over public company databases, exported to spreadsheets.

How we keep control

  • Isolation by organization from the first table, with row-level access control.
  • Figures calculated by explicit rules in code. The model does not make up a number.
  • The schema and the transformations are designed before the first system is connected.
  • Data and documents become a base an agent can query, with the source of every answer citable.
  • Every sync and pipeline logs what ran, what failed and when to retry.
  • Access is tested for each role and organization before we call it done.

How a project happens

Start with one task.
Grow by use.

  1. 01Scope one real task

    A concrete piece of work, with a start, an end and someone who does it by hand today.

  2. 02Build the base

    Data, authentication and the core rules come before the polish.

  3. 03Test on real use

    You follow the system in a test environment, review the flows and check answers against their sources.

  4. 04Ship and keep going

    Production with its own database, domain and monitoring. Evolution follows real use.

Questions

Before we start.

Do we have to buy the build from Tecela?

No. Advice and build are separate agreements. Any Tecela build proposal comes with a second option in writing, you own everything we make, and you can take the plan to any builder.

Which AI models do you use?

Whichever works best for the task. We work with models from OpenAI, Anthropic and Google, among others, and choose by test: quality of the answer, cost and speed. The choice is isolated in the code, so changing models later does not mean rewriting the system.

Do we have to replace our systems?

In many cases, no. We assess the APIs, permissions and limits of the tools you already use before we design the integration.

What if the AI is wrong?

Models can be wrong. That is why every answer points to a source you can check, with checks in code where they apply and human review before sensitive actions.

Where does our data live?

Access is separated by organization and by user, and an agent only reads what the task requires. We decide with you what may go to an external model and what stays only in your base.

What happens after the first version?

You receive the source code and the documentation. Hosting, maintenance and new features can continue with us, with scope, priorities and responsibilities defined with you from real use.

Next step

Which process should
run in production first?

Tell me how it works today and who needs access to what. We define a first slice from there.

Request a 30-minute call