Emmanuel Oliveira — Software Engineer

MentorGPT is a humanized LLM-based service platform created to allow companies to develop and configure custom AI agents from their own knowledge bases. The platform supported importing content in multiple formats, such as PDFs, documents, audio files, links, and authorized messages from multiple channels. These materials were processed and organized so each assistant could respond based on the context, data, and specific needs of each client. Each agent could be configured individually, and users could select different models and providers such as Gemini, GPT, and Groq, as well as define the system prompt, response temperature, access permissions, and the knowledge bases available to each agent. On the technical side, the solution used a hybrid RAG architecture, combining embedding search with graph-based knowledge structures stored in Neo4j. Document processing and other asynchronous tasks were handled with AWS SQS and AWS Lambda, providing better scalability and service decoupling. The system also offered integrations with Slack, chatbot embedding in authorized websites through iframe, two-factor authentication (2FA), and granular user, permission, and access management.

Python · Flask · AWS · AWS Cognito · AWS Lambda · AWS S3 · AWS EKS · AWS SQS · AWS SES · AWS ECR · OpenAI · Groq · Gemini · React + Vite · Slack

WhatsApp-based learning paths platform created to distribute multimedia content, track progress, and reinforce learning with gamification, spaced repetition, and LLM-powered question support. The system allowed creating and managing learning paths composed of videos, links, audio files, PDFs, and other materials stored in the cloud. For each piece of content, it was also possible to register related questions, and the materials and questions were processed and used as the LLM knowledge base. Users could start and complete the entire path directly through WhatsApp. After consuming each item, they received a configurable number of random questions related to the studied material. They could also ask questions and receive contextual answers based on the path materials. As they progressed, they earned points for completing content and answering correctly. The platform applied a spaced repetition strategy. After a piece of content was completed, the system scheduled new questions about previously studied topics, helping reinforce knowledge retention over time. One of the main concerns was preventing the contents from taking up space on participants’ phones. So a Rust-based link shortener was developed to direct users to a web interface, allowing content consumption without downloading files to the device. On the technical side, the solution was built with Python, Flask, Rust, and Tokio, using MongoDB for data persistence and Amazon S3 for content storage. Asynchronous processing was handled with AWS SQS and AWS Lambda. The platform also integrated with language models and the WhatsApp and Telegram APIs. Applied in the Rocinha community, the project received an international WITSA award in 2023.

Python · Flask · Rust · Tokio · MongoDB · AWS S3 · AWS SQS · AWS Lambda · WhatsApp API · Telegram API · LLMs

ZapVagas is a platform created to connect workers with job opportunities that usually do not appear on services such as LinkedIn or Gupy. The project focused on roles like waiter, baker, driver, attendant, and other positions with simpler and less bureaucratic hiring processes. For employers, organizations, or people responsible for publishing openings, the platform offered a simple and intuitive web interface. Jobs could be registered through a straightforward form containing the main vacancy details and the necessary contact information. For candidates, the entire experience happened through WhatsApp. Users only needed to send a keyword message, such as “I WANT TO RECEIVE JOBS”, to start receiving the openings registered on the platform directly in the app, without creating an account, installing another system, or filling out long forms. Each opportunity was sent in a well-structured message, with the information needed for the candidate to evaluate the role and contact the employer. The messages also included an unsubscribe option, allowing users to stop receiving new opportunities in a simple and transparent way. The solution was designed to be used by associations, NGOs, social projects, and city governments, making it easier to distribute local jobs and making the connection between employers and workers faster and more accessible. On the technical side, the platform was developed with Python and Flask on the backend, React and Vite on the frontend, and MongoDB for data persistence. The application ran on AWS using AWS App Runner.

Python · Flask · React + Vite · MongoDB · AWS App Runner · WhatsApp API

Access control system developed to manage the entry of students, teachers, staff, and visitors in academic environments such as classrooms, laboratories, libraries, and administrative areas. User identification was performed through cards or devices with RFID/NFC technology. When the credential was brought near the reader, the system queried a database-integrated API to verify the user identity and permissions before authorizing or denying access to the space. The platform allowed different user profiles to be configured and defined which locations each person could access, considering rules such as role, course, class, schedule, or authorization level. Each entry attempt could also be logged, allowing the access history and space usage to be reviewed. On the hardware side, the project used Arduino integrated with RFID/NFC readers to read credentials and control access devices. On the software side, the solution was developed in Python, with an API responsible for communication between the devices, the permission system, and the database. The solution was designed to improve security in educational environments, reduce manual checks, and centralize access management in a single platform.

Arduino · RFID/NFC · Python · API

stAIr was a pilot learning project created by MentorApp in partnership with the Brazilian Industrial Development Agency (ABDI) and the Ministry of Economy. The initiative was developed to add value to Brasil Mais, a free national program focused on increasing the productivity and competitiveness of Brazilian companies. The project aimed to train micro and small businesses in digital transformation, offering a practical, accessible learning experience tailored to each business need. The platform worked as a digital course environment, with an experience similar to streaming services. Participants could access learning paths, browse available content, and track their progress throughout the training. Each path was composed of structured materials about digital transformation. While consuming the content, users could also use an AI chat to ask questions and receive contextual answers based on the materials within the path. The proposal combined an organized on-demand learning experience with conversational support, allowing each participant to move at their own pace and consult the assistant whenever they needed clarification on course concepts. On the technical side, the solution was developed with Python and Flask, using Neo4j for knowledge organization and querying. The infrastructure ran on AWS, with Amazon EKS for service orchestration, Amazon S3 for content storage, and AWS Lambda for task processing. stAIr was one of the nine projects selected nationwide to support the spread of digital transformation among micro and small businesses.

Python · Flask · Neo4j · AWS EKS · AWS S3 · AWS Lambda

MentorApp Academy was a learning platform focused on developing the five main soft skills highlighted by Forbes. The solution brought together videos, articles, and other free materials in an experience similar to a streaming platform, allowing users to explore content according to their professional interests and needs. Without a fixed learning path, the platform offered a personalized experience through a recommendation system. With each interaction, similar content or content well rated by users with close profiles was automatically suggested. The engine used collaborative filtering with matrix factorization, based on the FunkSVD algorithm. From user-content interactions, the system identified patterns to estimate which materials would be more relevant to each person. The content library was updated automatically. Through web scraping, the system collected content from sites previously registered as trusted sources, and a machine learning model analyzed each item, classified the soft skill covered, and automatically published it in the corresponding category. Users could also rate materials and flag inappropriate, outdated, or low-quality content. On the technical side, the solution was developed with Python and Flask, using scikit-learn in the machine learning and recommendation models, Selenium for automated content collection, and MongoDB for data persistence. The infrastructure ran on AWS, with Amazon EC2 for hosting and AWS Lambda for automated tasks. MentorApp Academy thus maintained a dynamic and constantly evolving catalog, combining automated collection, intelligent classification, personalized recommendation, and community participation.

Python · Flask · Recommendation systems · scikit-learn · Selenium · MongoDB · AWS EC2 · AWS Lambda

The skill was developed to deliver interactive narrative experiences focused on business decision-making. Inspired by choice-based games with multiple paths, the solution allowed each user decision to change the course of the story and lead to different consequences and endings. Throughout the narrative, users faced strategic situations related to the business environment and had to choose between different alternatives. Each answer influenced the next events, creating a personalized journey and encouraging skills such as scenario analysis, risk evaluation, and decision-making. The stories were structured as branching flows, where different combinations of choices led to distinct paths and outcomes. This allowed users to repeat the experience, explore new decisions, and understand how each choice impacted the story development. At the end of each story, the skill presented feedback about the user performance, considering the decisions made throughout the experience. It was also possible to request a more detailed report by email, with an analysis of the chosen paths, the results achieved, and the points that could be improved. On the technical side, the solution was developed in Python and integrated with the Amazon Alexa platform. Interaction execution was handled with AWS Lambda, while Amazon S3 was used to store application content and assets. Neo4j was responsible for modeling the narratives, the relationships between decisions, and the different possible story paths.

Python · Amazon Alexa · Alexa Skills Kit · AWS Lambda · AWS S3 · Neo4j

MentorApp Alexa Smart Training was a skill created to provide quick productivity tips through Amazon’s Alexa voice assistant. The proposal worked as a daily tip. By using commands like “Alexa, how can I be more productive?”, the user received a short and practical orientation on topics such as organization, focus, time management, and personal and professional development. The experience was designed to make access to learning content simpler and more accessible, allowing users to consume small doses of knowledge during their routine without needing to access courses, apps, or long materials. On the technical side, the solution was developed in Python using the Alexa Skills Kit. Skill execution ran on AWS Lambda, while Amazon S3 was used to store content and DynamoDB to persist application data. Available for free in the Amazon store, the skill stood out by ranking among the 50 most downloaded in 2020.

Python · Alexa Skills Kit · AWS Lambda · AWS S3 · DynamoDB

Come Back Tomorrow was an educational prototype developed in Python and inspired by the behavior of ransomware such as WannaCry. The project was created during a technical course to demonstrate information security, cryptography, and digital threat behavior in a controlled environment. The application simulated file access blocking on Windows 7 and Windows 10 systems and presented the user with an interface similar to those used in ransomware attacks, including a fake ransom message and a deadline for data recovery. The project made it possible to study how this type of threat affects a system, how cryptographic mechanisms can be used maliciously, and which prevention, recovery, and protection measures should be adopted against attacks of this nature. All development was carried out exclusively for academic purposes, on test machines and files, without distribution or use in real environments. On the technical side, the solution was developed mainly in Python, involving file manipulation, creation of Windows interfaces, and application of basic cryptography and system security concepts.

Python · Windows · Cryptography