Generative AI for systems documentation and testing
Prodam's proposal is to use generative AI to speed up backlog, technical documentation and testing. The expected gain is less preparation time, better-organized context and a shorter path to execution.
Assistive AI
Probô was introduced as support for documentation and testing, focused on speeding up analysis, artifact writing and technical review.
Web + Azure DevOps
Microsoft sign-in and integration with wiki and work items whenever the context calls for traceability.
Gemini + Vertex AI
According to the institutional FAQ, the features use Gemini-family models on Google Vertex AI infrastructure.
From documentation to execution
Covers artifact generation, iterative refinement, test scripts, test data and repository analysis.
Product in focus
Documentation, backlog and tests generated with AI support
Early data model validation before submission to Data Administration
Conversational refinement to adjust results before publishing
Integration with wiki, work items and repository context
Operating principle
Repetitive steps get faster; critical steps still go through review, additional context and refinement before publishing.
Data model validation
Probô, an innovative Prodam application powered by generative AI, helps analysts and developers run more standardized checks early and catch inconsistencies before submitting the data model to Data Administration, delivering models better aligned with the standards and speeding up the process.
Before
The model went to Data Administration, came back flagged and started another round of fixes and resubmission.
With Probô
Standards are checked before submission. The model arrives better aligned and validation moves faster.
In ER/Studio, the export macro generates a CSV with the model's entities, attributes, data types, keys and definitions.
In Probô, under Analyze Data Model, attach the CSV, choose the project type and the modeling type, then click Process.
Findings are grouped by entity and attribute, separating what must be fixed from what is a recommendation.
Adjust the model in the modeling tool and ask the chatbot why each item was flagged before processing again.
Generate the PDF report and attach it to the request. The model goes to Data Administration, which still validates it.
Inconsistencies
Deviations from the modeling standards, such as plural names, missing definitions or non-standard abbreviations. Fix them before submitting.
Alerts
Recommendations that don't block submission, such as suggested indexes or data types to review. Weigh each one.
Validators still validate
Probô doesn't replace Data Administration: it checks the standards earlier so the official validation involves fewer round trips.
Always a human review. Like any generative AI, Probô can make mistakes. Review the generated content before using it.
Sign in with a Prodam Microsoft account.
Overview
Probô works as a work accelerator: it helps organize knowledge, automate parts of the cycle and shorten the path from understanding to execution.
What changes in practice
In a virtual session on Microsoft Teams, Carlos Eduardo Roque da Silva, Leonardo Ávila and Aline Souza de Araújo Santos walked through the use of AI in systems documentation and testing.
The presentation positions Probô as a tool to shorten documentation, backlog and testing steps while keeping technical review in the loop.
The flow combines preconfigured prompts, LLMs, additional context and iterative conversations to fit backlog, documentation and tests to the project's real scenario.
Expected benefits
Key message
“Probô organizes context, speeds up artifacts and shortens the cycle between analysis and testing.”
That sums up the positioning: practical use of AI in recurring tasks, built into the real engineering workflow.
Features
Rather than a loose list of prompts, Probô offers a set of flows applied to backlog, quality, documentation and systems analysis.
Generates features, user stories and epics from business requirements and turns use cases into a more actionable backlog.
Creates test plans, refines scenarios and produces automated scripts to speed up systems validation.
Uses institutional prompts and lets you create custom prompts with a goal, instructions, model and temperature.
Interprets legacy code, documents stored procedures and uses the repository wiki to answer questions about impact and behavior.
Requests can be enriched with text snippets, wiki links, PDFs, images and code paths.
Results can go to Word documents, wiki pages and Azure DevOps work items, keeping a trail of what was produced.
Flow
The value isn't only in generating content, but in adding context, reviewing and putting the output to work with a trail.
It starts with a prompt, a question or an artifact to generate, such as a feature, user story, test plan or SQL query.
Text, wiki, images, PDFs and code paths help the model answer with more precision and closer to the real scenario.
Use it iteratively: review, ask for changes, remove ambiguity and go deeper until the artifact is useful to the team.
After review, the content can be saved as a document, published to the wiki or turned into work items and runnable scripts.
Expected impact
The expectations for Probô go beyond a one-off gain. The tool is treated as a foundation for standardization, innovation and wider use of AI in adjacent areas.
Adoption horizon
The session closed with a path to expand Probô: more areas of the company, deeper insight into legacy systems, integration with other models and uses ever closer to the code and to real operations.
Strategic direction
Combining institutional knowledge, generative AI and integration with engineering tools creates a useful foundation to scale governance without making the flow heavier.
Next step
Take Probô beyond its initial context, broadening the tool's reach within the company.
Next step
Go deeper into the legacy context to support changes with a better understanding of dependencies.
Next step
Integrations with other generative AI players and models, opening room for the ecosystem to evolve technically.
Next step
Future plans include deeper source code exploration and studies on creating voice models.
FAQ
The main points of the institutional FAQ, focused on practical use, access, models, context and refining answers.
A generative AI application for software development. It supports analysis, documentation, backlog, artifact generation and testing with more speed and consistency.
The analyst exports the model's metadata from ER/Studio as CSV, attaches the file in Probô and gets a report by entity with inconsistencies and alerts. After fixing them, the PDF review is attached to the request. Data Administration remains responsible for the official validation.
The institutional FAQ highlights features, user stories and epics from business requirements, stored procedure documentation, user story review, test plans, Cypress and Playwright scripts, SQL queries in natural language and repository analysis.
It's a web application with Microsoft sign-in validated against Azure DevOps. Access through Prodam's Active Directory is also possible, with a narrower scope for features that depend on Azure DevOps.
The models help interpret text, generate documentation, suggest estimates, create scripts and analyze repositories. Use is iterative: the team asks, reviews, corrects and keeps what makes sense.
Prompts are the instructions given to the model. Temperature controls how precise or creative the answer will be. The tool offers preconfigured prompts and lets you create custom variations.
Gemini 2.0 Lite, 2.0 Flash, 2.0 Pro, 2.5 Flash-Lite, 2.5 Flash and 2.5 Pro, according to the institutional material.
Add free text, a wiki link, an image, a PDF or code paths to the request. The point is to reduce ambiguity so the model answers better and the team reworks less.
Because the best results come from interaction: review what was generated, ask for changes, provide examples and validate the artifact before publishing it or putting it to work.
The proposal, the features and where the tool is headed within PRODAM, in one page.
