BRS12 is a solo-builder AI product lab led by Rodrigo da Silva Souza, focused on building practical AI products for the Brazilian market using LLMs, agents, cloud infrastructure, and automation.
Rodrigo da Silva Souza is a Brazilian software architect and AI product builder with experience across public-sector portals, enterprise systems, open banking, education technology, APIs, microservices, frontend/backend development, Linux/server administration, and cloud-based product development.
His academic background combines postgraduate degrees in Software Architecture, Data Science & Big Data Analytics, and IT Project Management, plus degrees in IT Management and Pedagogy — a unique combination that informs both the technical depth and the educational focus of his AI products.
Public exams, government notices, and legal documents are dense, bureaucratic, and hard to study. Millions of candidates need structured, AI-processed study materials.
Brazilian financial market data (B3, stocks, options) is scattered across sources. Traders and analysts need AI assistants that understand local context.
Most AI tools are designed for English-speaking markets. Brazilian Portuguese, local data formats, and regional business workflows demand specialized adaptation.
BRS12 builds practical AI products that combine LLMs, agents, structured data pipelines, document understanding, chat interfaces, automation, and cloud infrastructure — all adapted for Brazilian Portuguese and local market needs.
AI education platform that transforms long official exam notices into structured summaries, study plans, flashcards, learning pills, and contextual chat experiences for Brazilian public exam candidates.
AI-assisted learning platform for ENEM, Brazil's national high school exam. Simplifies study paths, revision, practice, and learning engagement for one of the country's main university admission paths.
AI assistant for Brazilian financial market data, B3 workflows, quotes, options, market data integrations, and AI-assisted analysis. Focus on Brazilian stock market intelligence.
Agreement and trust-flow platform for transactions, validation, payment, and delivery between parties. AI can assist with agreement review, risk analysis, summarization, and workflow automation.
SaaS for YouTube transcription, clipping, and content processing workflows. Uses AI to process long-form video content into structured text, summaries, clips, and reusable knowledge assets.
Internal and public tools around LLM gateways, LiteLLM-based infrastructure, CLI tools, Hermes Agent workflows, MCP servers, AI memory infrastructure, and observability for model usage — monitoring token usage, costs, and anomalies.
Processing complex legal, regulatory, and educational documents into structured knowledge — summaries, flashcards, study plans, and searchable indexes.
Deep integration of LLMs into the development workflow: code generation, review, debugging, architecture design, and full-stack product building with AI copilots.
Hermes Agent and multi-agent orchestration for complex tasks — code review pipelines, multi-model comparison, automated testing, and structured output verification.
Multi-model architecture comparing Qwen, DeepSeek, Kimi, GLM, MiniMax, and Claude across tasks — routing to the best model for each workload.
JSON-mode output for study plans, flashcards, summaries, and data pipelines — reliable structured data from unstructured inputs.
AI systems that understand B3 tickers, options chains, market data formats, and Brazilian financial workflows — providing analysis and insights.
AI pipelines that convert long-form video and audio into structured text, summaries, clips, and searchable knowledge bases.
Planned specialization of AI models for Brazilian Portuguese public exam preparation, legal documents, and educational content — building on accumulated domain data.
BRS12 adopted Alibaba Cloud as one of its reference AI providers because of its cost/performance, model availability, and delivery quality. Alibaba Cloud has been effective for development with Qwen and other model families, and the next step is to scale usage through startup support, model credits, infrastructure credits, and technical guidance.
Reasoning, delegation, and general agent workflows
Coding, software development, and large-context workflows
Model comparison, structured generation, and routing
Future visual generation and educational asset creation
APIs, agents, databases, object storage, and async job processing
Scaled access to Qwen, DeepSeek, Kimi, GLM, MiniMax, and visual models for growing product usage and experimentation.
Compute resources for APIs, agent execution, async job workers, and production-grade service hosting.
Persistent storage for structured educational data, market data, user content, and document processing pipelines.
Job queues and async workers for document processing, content extraction, and AI pipeline orchestration at scale.
Semantic search infrastructure for educational content, document retrieval, and knowledge base queries.
Training infrastructure and guidance for fine-tuning models specialized in Brazilian Portuguese education and legal domains.
Tech guidance: Production deployment best practices, architecture reviews, and cloud optimization for AI-native workloads.
I'm building practical AI products for the Brazilian market — education, financial data, developer tooling, and automation. If you're from a startup program, cloud provider, or interested in collaborating, let's talk.