AI for public tenders, built for the whole process.
Not a general-purpose model with a specification pasted in. Carabela knows your company, reads the specifications as requirements, drafts against the award criteria, scores its own draft and learns from the result.
Anyone can paste a specification into a general-purpose AI model and ask for a summary. What comes back is reasonable text. What does not come back is what you actually need in order to bid: which requirements are mandatory and which earn points, what solvency you have to prove, which annexes have a prescribed format, how the points are split across the award criteria and what could get your bid excluded. That model does not know your company, does not know what you have delivered before and forgets the tender the moment you close the window.
Carabela is AI for public tenders built around the process. Before it reads a single specification, it builds a picture of your company: what you sell, what you have done, what you can prove and which contracts you want to go after. With that context, the tender analysis turns hundreds of pages into structured requirements, the drafting is done against the award criteria rather than against a blank page, and every claim in the technical proposal rests on real information about the company. You decide; the operational work belongs to the system.
And it does not depend on a single model. Carabela works with several AI models and picks the most suitable one for each task: one to read documentation, another to reason about fit, another to draft. When a better one appears, it is brought in. What does not change is what it is asked to do: read the whole file, write only what can be backed up, and learn from every result so that the next bid starts from a better position.
What the AI does inside the tender file
Builds a picture of your company
Before it searches for or drafts anything, Carabela learns what your company sells, which services it provides, which projects it has delivered, what experience and certifications it can prove, which requirements it can meet and which it cannot, and what kind of contracts it wants to go after. It can start from partial information. That context is what separates a search with judgement from a list of alerts, and a proposal of your own from generic text.
Turns hundreds of pages into structured requirements
It reads the PCAP, the PPT and the annexes and pulls out everything that has to be answered: subject matter and lots, budget and duration, deadlines, award criteria and price formulas, technical requirements, solvency, required accreditations, guarantees, page limits, annexes with a prescribed format and potential grounds for exclusion. Requirements that need review or confirmation from your company are flagged from the start.
Drafts against the award criteria
Before writing, it works out what the contracting body values and how the points are split. Then it allocates the content according to that scoring: methodology, work plan, team, schedule, resources, quality, risks, improvements and commitments. Every claim rests on real information about your company, and the system avoids inventing capabilities that cannot be backed up. It is the same logic it uses to prepare the complete technical proposal.
Scores its own draft and rewrites whatever loses points
Before submission, Carabela reviews the whole bid against the tender's scoring scheme: unanswered requirements, contradictions between the technical proposal, the price offer and the administrative documentation, incomplete annexes, page limits exceeded, unsupported information, award criteria only half covered and risks of exclusion. Whatever lowers the score is rewritten before it reaches your desk.
Learns from the result, not only from the file
When the award is published, Carabela analyses whatever is available: the outcome, the score obtained, the competitors' scores, the winning bid, the price differences and the technical assessments. Out of that come the weak points and what worked. It goes back into the system, so the next tender does not start from scratch: it starts from what was learnt on the previous one.
Picks the AI model to suit the task
Reading documentation, reasoning about fit, generating a long document and checking consistency are not the same task, and no single model does them all equally well. Carabela works with several AI models and uses the most suitable one for each part of the process. When a better capability appears, it is brought in. You do not depend on a single provider or a single AI.
How it works, step by step
It is the same sequence you can watch in [the whole system](/producto), seen from what the AI does at each stage.
- 01
It gets to know your company
Carabela builds a picture of your organisation: what you sell, what you have delivered, what you can prove, what resources you have and which contracts you want to go after. That is what lets it judge which tenders are worth pursuing and what material the bid can be written from.
- 02
It reads the specifications in full
The PCAP, the PPT and the annexes become structured requirements: mandatory, scored and formal, each with its deadlines, its solvency and its price formulas. The ones that need your company to confirm them are flagged from the start.
- 03
It drafts against the criteria
With the points allocated per criterion, the technical proposal is structured around what is assessed, and the administrative and price parts are prepared alongside it and cross-referenced. Every claim rests on what your company can back up.
- 04
It scores itself
Before the bid reaches you, Carabela reviews it in full against the scoring scheme: what is missing, what contradicts itself, what is not backed up and what covers a criterion only halfway. Whatever loses points is rewritten.
- 05
It learns from the result
Once the award is published, the system analyses your score, the competitors' scores, the winning bid and the price differences, where they are available. What was learnt goes back into your company's context for the next tender.
Who it is for
For companies that sell, or want to sell, to public bodies, in any sector.
- Companies that want to start selling to the public sector and need to know which opportunities exist and which requirements are usually asked for.
- Companies that already bid but submit few bids because there is no time to read specifications and prepare documentation.
- Companies that find opportunities and do not know which ones to prioritise.
- Bid teams that are small for the volume of tenders in their sector.
- Companies whose knowledge is spread across folders, documents and people, and that want to professionalise the process.
- Tender departments that want an extra layer of capacity so their people can concentrate on strategic decisions and exceptions.
Where you stay in charge
You choose which decisions Carabela takes on its own and which wait for your approval. Some you can always keep for yourself: bidding for or discarding a tender, requesting clarifications from the contracting body, setting the final price and submitting the proposal. You can approve each opportunity before a bid is prepared, review only the significant financial decisions, validate specific commitments before they go into the proposal, or delegate almost everything and receive only the exceptions. The level of autonomy is configured for each company, and nothing is submitted without you.
Frequently asked questions
How is this different from pasting the specification into a general-purpose AI model?
A general model can help you read or draft a text now and then. It does not know your company, does not know what you have delivered, cannot tell a mandatory requirement from a scored one and does not remember the last tender. Carabela brings together the company's context, the analysis of the file, the award criteria, the documentation, the strategy, the preparation of the bid, the follow-up and the learning in a single system. The difference is not the model: it is everything around it.
Which AI models does Carabela use?
Several, and not always the same ones. Carabela selects the most suitable model for each task: reading documentation, reasoning about fit, generating documents, checking consistency. It does not depend on a single provider. Because the technology moves quickly, the system is under continuous development and brings in new models and capabilities when they add real value. What stays constant is the process and the control you have over it.
Does the AI invent capabilities my company does not have?
It is designed not to. Carabela drafts from the tender's criteria, your company's documentation and its real capabilities, and aims for every claim in the proposal to match the information available. The pre-submission review flags unsupported information, and requirements that need confirmation from you are marked so that you can validate them before they go into the proposal.
Can it read any specification and any annex?
Carabela analyses the tender documentation, including the main documents and the annexes available: the PCAP, the PPT, the annex templates and the documentation published during the process. From those it extracts requirements, criteria, dates, conditions, required documentation and risks. The exact requirements always depend on the tender, and the ones that need interpretation or confirmation from your company are flagged for you to review.
Do we have to change the way we work?
Not necessarily. Carabela adapts to each organisation's workflow: how opportunities are reviewed, who approves each decision, which documentation is used, how proposals are produced and at which points a person steps in. If your company has its own processes, sources of information or approval flows, the system can be configured around them.
Does it guarantee that we will win?
No. Nobody can guarantee that a public body will award a particular contract: the decision rests with the contracting body and depends on the bids submitted. What Carabela does is improve how opportunities are selected, the quality and consistency of the bids, compliance with the requirements and your company's capacity to bid for more tenders that are a good fit.
Other solutions
All solutions →Watch the AI work on a real tender.
Book a demo with a real tender on the table and see what it reads, what it drafts and where it stops to ask you.