Name in original language
Ulysses - Roteador de demandas legislativas
Initiative overview
An AI system developed by the Brazilian Chamber of Deputies to automate the distribution of legislative demands from parliamentarians to specialized consultancy groups. When Members of Parliament (MPs) submit requests for bill drafting or research support, the system employs semantic analysis to interpret the content and thematic complexity of each demand, subsequently recommending up to three expert groups best suited to fulfill the request. A legislative consultancy staff member reviews these AI-generated suggestions before authorizing distribution, though demands with confidence scores exceeding 90% are routed automatically.
The solution pursues three main objectives:
- Enhance operational efficiency by accelerating the demand distribution process and reducing processing time
- Minimize allocation bias by applying consistent, data-driven criteria rather than subjective human judgment
- Reduce administrative burden on legislative consultancy staff, allowing them to focus on higher-value supervisory tasks rather than manual classification work.
The system leverages historical demand distribution data for model training and integrates with existing digital platforms used by MPs and legislative consultants. Success is measured primarily through Top-1 accuracy rates, assessing the percentage of demands correctly classified by the AI model. Key implementation challenges include maintaining model accuracy over time, adapting to evolving distribution rules, and monitoring organizational changes within lawmaking expert groups
Results, outcomes and impacts
- Decrease the staff work for reading and interpretating the demand texts
- Decrease the likehood of biases for interpeting and choosing the group of experts to be in charge of the new bill drafting
- Improved the incorporation of new subjects in the bills, especially when there is a blend of many subjects that would encompass many groups of lawmaking experts
Challenges and lessons learned
- To substantially enhance the performance of the AI model, it was necessary to join many legislative experts (from differrent groups of experts) to analyse historical data, which required a deep legislative knowledge, before the AI training.
- As demand distribution rules change over time, it was required to discard obsolete training data



























