An artificial intelligence (AI) tool developed last year by São Paulo-based startup NeuralMind promises to expedite procurement fraud investigations at the Brazilian Audit Court (TCU). The tool—called “AI-Powered Discovery” (Instrução Assistida com Inteligência Artificial, or INACIA)—is a search engine for mining data from legal documents in Portuguese using OpenAI’s GPT-4 language model.
INACIA is designed to assist TCU auditors in extracting meaningful information when reviewing ongoing cases. “The system enables auditors to quickly understand the core facts of case, compile available evidence and identify relevant legal precedents,” explains Roberto Lotufo, a computer engineer and cofounder of NeuralMind, which operates out of the Science and Technology Park at the University of Campinas (UNICAMP). The firm’s NeuralSearchX software, built on neural network technology, enables more accurate searches across large collections of Portuguese-language documents. The initiative was featured in a March 2024 article in Digital Government: Research and Practice.
NeuralSearchX is one of 383 AI-based projects that have received grant funding from FAPESP’s Innovative Research in Small Businesses (PIPE) program over the past 26 years. The number of AI-related projects within the program has grown exponentially over this period, with a 20-fold increase in the last decade—from just three AI projects and R$245,000 in grant funding in 2013 to 61 projects and nearly R$12.8 million in funding in 2023 (see chart). Since PIPE’s launch in 1997, AI projects have been primarily concentrated in exact and Earth sciences (43%), followed by engineering (23%) and agricultural sciences (12%).
“AI is the tech wave of the moment,” says Rodolfo Azevedo, a computer scientist who heads the Technologies and Innovation Partnerships department at FAPESP and is a professor at the Institute of Computing at UNICAMP. FAPESP, he explains, is funneling much of its grants to projects developing new ways to leverage these technologies. “It’s also important that startups focus their innovation efforts on lowering costs and making processes scalable. By doing so, they are more likely to make a meaningful impact on people’s lives,” says Azevedo.
He notes that many startups follow a similar path of using well-established architectures or algorithms and extracting new insights by applying them to proprietary datasets or integrating new solutions, like NeuralMind’s search engine, which integrates with GPT. “Companies typically spend most of the PIPE grant period training personnel in AI, a field in which there is still a very limited skilled workforce, as well as fine-tuning AI models for specific applications,” he adds.
“The growth seen in recent years mirrors the broader expansion of AI startups in Brazil and globally,” notes Eduardo de Rezende Francisco, a computer scientist at Fundação Getulio Vargas (FGV) in São Paulo. According to the Emerging Tech Report 2024 from Brazilian-based startup platform Distrito, 45% of the 2,252 emerging tech startups in Latin America are focused on AI, followed by the Internet of Things (IoT) at 16%, and datafication—the conversion of information into digital data—at 12%. Brazil leads the Latin American AI market with 74% of the region’s AI startups.
Latin American startups exploring emerging technologies have attracted US$3.8 billion in seed funding, according to the report, with the AI sector accounting for 60% of this at US$2.4 billion. “Nonrepayable grant programs like PIPE play a key role in funding research and innovation with less constraints, especially in the field of AI,” Francisco notes.
Globally, the AI market was valued at US$196.6 billion in 2023, according to market intelligence firm Grand View Research, with the dominant players being the US, China, the European Union, and the UK. In 2022, private investment in AI reached US$92 billion—18 times higher than in 2013—as per the AI Index Report 2023 from Stanford’s Institute for Human-Centered Artificial Intelligence (HAI).
Tracking sleep and climate
Vigilantes do Sono, a startup incubated at São Paulo’s Hospital Israelita Albert Einstein, received FAPESP funding to create an app that uses machine learning to suggest strategies for improving sleep quality. Its main features include an insomnia treatment protocol based on cognitive-behavioral therapy and a screening tool for sleep apnea.
The insomnia section features chatbot-guided conversations, while a sleep apnea test uses smartphone sensors—such as the camera, accelerometer, and microphone—to detect physical and behavioral signs of the disorder. “Our goal is to make sleep health more accessible, especially in Brazil, where access to treatment and specialized labs is limited. Eventually, we hope to integrate with the national healthcare system (SUS),” explains psychologist Laura Castro, a cofounder and currently head of psychology and research at the company.
The algorithm for the sleep apnea module—the latest feature on the app—was developed, trained, and validated as part of an ongoing Phase 2 PIPE project. The goal is to help detect symptoms indicating a potential risk of developing the disorder; based on the results, the tool may suggest the user seek specialized medical attention, for instance.
Meanwhile, São Paulo-based startup MeteoIA, another PIPE grant recipient, developed a neural network-based system to generate customized models for predicting long-term climate risks. These models have a wide range of applications, such as forecasting landslides on highway slopes or predicting climate anomalies that could affect agricultural yields.
In the energy sector, MeteoIA created a model for hydroelectric plants, trained as part of two PIPE projects between November 2020 and April 2024, to forecast rainfall in watersheds hosting hydroelectric dams within the National Grid (SIN). “The system forecasts rainfall anomalies over the next 12 months with 70% accuracy. It doesn’t predict the exact day the rainfall will occur, but it identifies the months when higher-than-expected precipitation is more likely,” explains Thomas Martin, a scientist and cofounder of the startup.
The story above was published with the title “Innovative applications” in issue 344 of October/2024.
Projects
1. Validation of NeuralSearchX for searching tax databases and integration with MS Office (n° 22/13727-5); Grant Mechanism Innovative Research in Small Businesses (PIPE); Principal Investigator Roberto de Alencar Lotufo (NeuralMind); Investment R$431,363.96.
2. QUEST ‒ System for searching and aggregating information based on Zero-Shot techniques (n° 22/01640-2); Grant Mechanism Innovative Research in Small Businesses (PIPE); Principal Investigator Rodrigo Frassetto Nogueira (NeuralMind); Investment R$1,048,816.23.
3. Intelligent system for case law analysis using modern deep learning with natural language processing (n° 20/09753-5); Grant Mechanism Innovative Research in Small Businesses (PIPE); Principal Investigator Rodrigo Frassetto Nogueira; Investment R$431,078.52.
4. Robust reading of registration documents using deep learning (n° 19/06667-3); Grant Mechanism Innovative Research in Small Businesses (PIPE); Principal Investigator Roberto de Alencar Lotufo (NeuralMind); Agreement Sebrae-SP; Investment R$662,802.81.
5. System for robust reading of text in images using deep learning (n° 18/01188-7); Grant Mechanism Innovative Research in Small Businesses (PIPE); Principal Investigator Roberto de Alencar Lotufo (NeuralMind); Investment R$410,479.61.
6. Intelligent phenotyping with machine learning for therapeutic personalization in digital sleep improvement program (n° 21/12139-0); Grant Mechanism Innovative Research in Small Businesses (PIPE); Principal Investigator Laura de Siqueira Castro (Vigilantes do Sono); Investment R$957,782.71.
7. MIA: Artificial intelligence system for subseasonal forecasting of hydrometeorological variables in Brazil (n° 20/00566-8); Grant Mechanism Innovative Research in Small Businesses (PIPE); Principal Investigator Thomas Christian Marcel Martin (Meteoia Datascience); Investment R$457,307.60.
8. MIA: Artificial intelligence system for subseasonal forecasting of hydrometeorological variables in Brazil (n° 21/14700-0); Grant Mechanism Innovative Research in Small Businesses (PIPE); Principal Investigator Thomas Christian Marcel Martin (Meteoia Datascience); Investment R$583,435.78.
Scientific article
PEREIRA et al. Inacia: Integrating large language models in Brazilian audit courts: Opportunities and challenges. Digital Government: Research and Practice. Mar. 2024.
Republish
