# Mazetech.ai — Full content for AI agents

> Artificial intelligence applied to corporate processes of the highest complexity. Founded and operated in Brazil. 9 years in market, 4 sectors served, 1M+ high-complexity items processed in production.

## Approach

We work where critical processes demand more than a plug-in model: regulation, contracts, unstructured documents, human exceptions, and decisions with financial impact. We dive into the partner company's target process until we understand it as well as the people who run it. From there we design the entire application — front-end to OCR microservices, embedding models, LLM agents, and proprietary review mechanisms.

- **Process immersion** — We map the real flow, exceptions, human actors, and decision points before writing a single line of code.
- **Systems, not scripts** — Senior teams shipping production products, not disposable MVPs. Interfaces, distributed back-ends, OCR pipelines, RAG, embeddings, agents, and proprietary models.
- **Proprietary technology** — The IP of what we build is ours. Proven with a reference partner, then scaled across the entire sector.

## Business model

We are not a software house. We build proprietary technology to solve complex processes — and turn those solutions into products for entire sectors.

- **Partnership, not outsourcing** — We don't sell hours or build-on-demand squads. We come in as a long-term tech partner. Result: a mature, production-ready solution, in less time, with less risk, and for a fraction of the cost of other vendors.
- **Vertical depth, scalable product** — We pursue specific, recurring problems where simple automation isn't enough. Each deployment sharpens the product, the models, and the sector knowledge. Every partner benefits from what we learn with the others, including the first one.

## Products in production

### Edupass — Higher Education
Automation of academic transfers. Advanced OCR for transcript extraction, NER and embeddings to structure courses, NLP for exemption suggestions, with proprietary LLM agents reviewing the process.
- Tech: Advanced OCR, NLP, NER, Embeddings, LLM agents, Pipeline at scale.
- Partners: Cogna, Ânima, YDUQS.
- URL: https://edupass.com.br

### Plataforma Blox — Higher Education
End-to-end automation of academic flexibility in higher education. Machine-learning recommender models help each student find the best courses for their academic moment; linear-optimization algorithms maximize the operational efficiency of students per classroom.
- Tech: Recommender System, Linear Optimization, Pipeline at scale.
- Partners: Higher education institutions.
- URL: https://plataformablox.com.br

### Intellihealth — Hospital Care
Automated defense of hospital claim denials. Searches contractual bases for legal-technical grounds, building appeals with RAG, embedding models, and specialized agents.
- Tech: RAG, Embeddings, LLM agents, Document automation, Pipeline at scale.
- Stage: Proof of Concept (PoC) — validating the technical approach and real-world impact at reduced scale before full operation.
- Partner: Hospital Sírio-Libanês.

### Intellidocs — Labor Law
Processes thousands of labor lawsuits extracting relevant information and entities (NER), embeddings for RAG, and regression models that estimate the likely return value of each case.
- Tech: NER, RAG, Regression models, Pipeline at scale.
- Partners: Corporate legal departments.

### Intellibook — Publishers
Automatic book ingestion with metadata extraction — chapters, table of contents, entities. Lets readers chat with the book through any messaging tool, including WhatsApp, using RAG and proprietary agents.
- Tech: NLP, Embeddings, RAG, LLM agents, Omnichannel chat, Pipeline at scale.
- Partner: Editora do Brasil.

## Technology stack

We build and operate each application end-to-end — front-end to models, through architecture and DevOps. Clients consume it as SaaS.

- **LLM Models** — Proprietary agents.
- **ML Models** — Clustering, regression, classification.
- **NLP** — NER and classification.
- **Computer Vision** — Detection and image classification.
- **Retrieval** — RAG and Embeddings.
- **Documents** — Advanced OCR.
- **Recommendation** — ML-based recommender systems.
- **Optimization** — Linear programming and constraint solving.
- **Architecture** — Monolith + microservices.
- **Front-end** — Web apps or chat interfaces.
- **BI** — Dashboards and visualizations.
- **SaaS Operations** — DevOps, scale, observability.

## Glossary

- **OCR** (Optical Character Recognition) — Advanced pipelines that extract text, tables, and structure from scanned documents, even at poor quality.
- **Computer Vision** — Models that extract information from images: object detection, visual classification, segmentation. Used to read documents, identify patterns, and automate inspections.
- **NLP** (Natural Language Processing) — Computational analysis of language for classification, extraction, and structuring of unstructured text.
- **Embeddings** — Numerical (vector) representations of text. Allow searching and comparing passages by meaning, not just by words.
- **LLM Agents** — Agents built on Large Language Models — orchestrations that reason, decide, and execute complex tasks under proprietary review policies.
- **RAG** (Retrieval-Augmented Generation) — Combines search over proprietary corpora with language generation, producing precise answers anchored in real documents.
- **NER** (Named Entity Recognition) — Identifies and structures names, values, dates, case parties, and other entities in free text.
- **Regression Models** — Statistical and ML models that estimate numerical values — for example, the likely return value of a lawsuit or the risk of a transaction.
- **Recommender System** — Machine learning models that suggest items (courses, products, content) by learning from user history and context to nail the next choice.
- **Linear Optimization** — Algorithms that find the best possible allocation given an objective function and a set of constraints — e.g., distributing students across classrooms to maximize occupancy while respecting academic rules.
- **Document Automation** — Full pipelines that process documents at scale: reading, classification, extraction, validation, and composition of new documents.
- **Pipeline** — Processing infrastructure running at industrial scale — thousands of items processed reliably, with monitoring and auditability.
- **Chat Channels** — Integration with WhatsApp, Telegram, and other chat channels. Users converse with the product through the tools they already use every day.
- **Architecture** — Solid monolith for CRUD and orchestration; microservices for components that need to scale independently or run on specialized hardware.
- **ML Models** — Classic machine learning models — clustering, regression, classification — for prediction, segmentation, and estimation. Combined with LLMs where it makes sense.
- **BI** — Tailor-made BI: dashboards and analytical visualizations wired directly to production data. The indicators that matter for each operation, not generic reports.
- **SaaS Operations** — Clients consume it as SaaS. We handle DevOps, deployment, observability, security, and scale — no infra provisioning, no SRE hires, no incident operations on their side.
- **Front-end** — Tailor-made web applications or interfaces inside chat tools the user already uses — WhatsApp, Telegram, Slack — integrated directly with the back-end models and pipelines.
- **In Production** — System running live with real clients in real environments for over a year.
- **Proof of Concept (PoC)** — A working version of the system validating the technical approach and real-world impact at reduced scale, before full operation with clients.

## What we evaluate in a partnership

- Processes with significant volume and documental complexity.
- Relevant, established companies in the target sector.
- Sector-wide scale potential.

## Contact

- Email: contato@mazetech.ai
- Address: Cubo Itaú — Al. Vicente Pizon, 54, Vila Olímpia, São Paulo, Brazil
- Maps: https://maps.app.goo.gl/cZoSnWKKRU3KKriDA

## Versions

- Portuguese (default): https://mazetech.ai/
- English: https://mazetech.ai/en/
- Spanish: https://mazetech.ai/es/