João José Bentivi
Verified Expert in Engineering
Full-stack Software Engineer and AI Systems Developer
São Luís - State of Maranhão, Brazil
Toptal member since May 15, 2026
João is a full-stack software engineer and AI systems developer with 10+ years building production-grade SaaS and AI platforms. His expertise spans Next.js, React, TypeScript, Supabase/PostgreSQL, OpenAI Responses and Realtime APIs, RAG, multi-agent systems, multi-provider LLM orchestration, and idempotent credit and billing systems. Founder of DreamSteps, Simula IA, and Pensador, and former head of engineering and infrastructure at MackAI, João bridges applied research and product strategy.
Portfolio
Experience
- Python - 12 years
- Product Planning - 8 years
- RAG Architecture - 4 years
- APIs - 4 years
- LangChain - 4 years
- LLM Integration - 3 years
- Prompt Engineering - 3 years
- Model Context Protocol (MCP) - 1 year
Preferred Environment
Linux, Python, LangChain, RAG Architecture, LLM Integration, Prompt Engineering, SQL, Model Context Protocol (MCP), Amazon EC2, APIs
The most amazing...
...thing I've developed is JurisAI, a legal full-stack AI SaaS with RAG, HNSW, streaming chat, billing, MCPs, and multi-agent reasoning.
Work Experience
Founder & Lead Engineer
DreamSteps
- Architected and shipped DreamSteps, a bilingual (pt-BR/en) AI business-consulting SaaS built on Next.js 16 (App Router), React 19, TypeScript, Tailwind v4, and Supabase/PostgreSQL, with 22 API route handlers.
- Built a staged AI interview engine: 6 sequential interviews plus 3 derived reports, with adaptive follow-up questions, per-attempt language locking, resumable sessions, and database-first bilingual content editable from an admin console.
- Designed a billing contract for paid AI operations: client idempotency keys, durable operations claimed through SECURITY DEFINER Postgres RPCs with fencing tokens, and credit reservations.
- Implemented a provider-agnostic LLM layer over OpenAI, DeepSeek, and OpenRouter with per-plan authorization of provider, model, and reasoning effort; and untrusted-input wrapping against prompt injection.
- Delivered monetization and operations: Stripe subscriptions (Free/Pro/Max) with weekly credit allowances and out-of-order-safe webhook processing, a 5-persona AI advisory board with board-level synthesis.
- Implemented transactional and marketing email via Resend/React Email, including signature-verified inbound mail, support ticketing, and an admin console for plans, providers, content, and usage.
- Hardened and deployed the platform on AWS EC2 behind nginx and Cloudflare (PM2, Origin CA): per-request CSP nonces, same-origin enforcement, per-IP and per-user rate limiting, bounded JSON parsing, and Postgres Row-Level Security.
- Contributed to Cloudflare Access JWT for admin, fail-closed boot checks for env and schema drift, and a 20+ suite Vitest security contract test suite.
AI Chatbot & Automation Engineer
Dance With Me
- Built a lead-capture and profiling flow that collected website visitor information, class interests, preferences, and user profile data, storing structured records in a centralized database.
- Developed an AI-powered chatbot system for Dance With Me, a dance school, to automate lead nurturing, client communication, and front-desk support workflows.
- Implemented automated nurturing sequences with timed follow-ups, including 1-day, 2-day, and 5-day message intervals, to engage prospective students with relevant school, class, and enrollment information.
- Designed personalized chatbot conversations that adapted responses based on each lead’s interests, dance preferences, goals, objections, and engagement history.
- Integrated the chatbot with school databases through MCP-based tool workflows, enabling the system to retrieve student-specific information such as upcoming classes, schedules, enrollment details, and payment-related data.
- Developed a multi-agent architecture with orchestration, validation, intent-analysis, and task-specialized agents to handle diverse user requests across sales, support, scheduling, and student service scenarios.
- Automated repetitive front-desk interactions by enabling students and leads to receive personalized answers about classes, plans, school activities, and administrative information without requiring manual staff intervention.
- Improved lead management and customer support scalability by organizing prospect and student data into actionable conversational workflows for follow-up, qualification, and retention.
Senior Full-stack Developer
Literatura Viva/Pensador
- Architected and shipped a production full-stack AI SaaS platform using Next.js 16, TypeScript, Supabase/PostgreSQL, and OpenAI Responses API, supporting conversational AI, research workflows, user authentication, and subscription-based access.
- Built a real-time streaming AI chat engine using the OpenAI Responses API, enabling tool calling, reasoning-summary parsing, persistent conversation history, and low-friction user interaction.
- Developed a Python MCP server exposing author-specific RAG retrieval tools across multiple philosophical corpora, allowing users to query curated knowledge bases through structured AI tools.
- Implemented a semantic search infrastructure using OpenAI text-embedding-3-large and HNSW/hnswlib, achieving sub-millisecond approximate-nearest-neighbor retrieval across philosophical text corpora.
- Created a parallel multi-agent analysis framework based on Edward de Bono’s Six Thinking Hats methodology, executing 6 specialized agents concurrently to generate a multi-perspective research synthesis.
- Built an asynchronous worker pipeline for long-running AI tools, including queued requests, processing states, timeout guards, external API execution, persisted results, and front-end polling for completion.
- Implemented security and access-control foundations using Supabase Auth, Google OAuth, server-side cookies, PostgreSQL Row Level Security, protected API routes, secret-based worker authentication, and server-side validation.
- Established product observability through structured API logs, redacted sensitive fields, tool invocation tracking, usage analytics, latency aggregates, costs, and admin-facing performance monitoring.
Senior Full-stack Developer
JurisAI
- Designed and led the product development of JurisAI, an AI-driven legal research platform that reached 600+ active users and supported lawyers with context-aware legal research workflows.
- Built a RAG-based legal retrieval system that enabled users to search, analyze, and discuss legal questions using laws, jurisprudence, and legal doctrine as contextual knowledge sources.
- Developed intelligent legal workflows for document generation, legal argument refinement, and ingestion of procedural documents across multiple file formats.
- Scaled the platform to more than 2,000 legal queries, validating the product’s practical use among legal professionals and law-firm workflows.
- Collaborated with legal experts and stakeholders to translate domain-specific legal requirements into AI workflows, retrieval logic, and product features.
- Created a specialized legal AI assistant that helped lawyers explore legal theses, refine arguments, and interact with complex legal knowledge beyond generic chatbot capabilities.
Senior Full-stack Developer
Luminus Solucoes em Inteligencia Artificial
- Designed and led the engineering of JurisAI, an AI-driven legal research platform with 700+ active users.
- Built an LLM-based retrieval system using RAG architecture for context-aware legal information retrieval.
- Orchestrated model fine-tuning and continuous-learning pipelines to improve answer relevance and performance.
- Established scalable cloud infrastructure on AWS, including containerization (Docker) and CI/CD pipelines.
- Collaborated with legal experts and stakeholders to embed domain knowledge into AI workflows.
Head of Engineering & Infrastructure
Mack ai
- Planned, designed, and deployed scalable server infrastructures supporting multiple mission-critical applications.
- Established an organizational culture focused on collaboration, innovation, and continuous improvement.
- Coordinated onboarding and technical training programs for new engineering hires, accelerating time-to-productivity.
- Implemented end-to-end DevOps pipelines, CI/CD, containerization, and automated testing to streamline deployments.
Independent Software & Machine Learning Developer
Freelance Clients
- Developed Python-based machine learning projects spanning recommendation systems, predictive modeling, natural language processing, clustering, and deep learning.
- Built a predictive recommendation system and experimented with supervised and unsupervised learning techniques using real-world datasets.
- Implemented deep learning models with Keras and explored predictive modeling for automotive sales data.
- Applied NLP techniques to text analysis and used K-Means clustering for unsupervised pattern discovery.
- Used Python, Jupyter Notebook, Pandas, NumPy, Scikit-learn, Keras, and data visualization libraries for model development and exploratory analysis.
Experience
DreamSteps
https://dreamsteps.org/enI own the product and the entire codebase: Next.js 16, React 19, TypeScript, and Tailwind v4 on the front end; Supabase/PostgreSQL with Row-Level Security and SECURITY DEFINER RPCs on the back end; Stripe subscriptions with a weekly credit system for monetization.
The hardest problem was billing correctness, since every AI call costs real money. I designed an exactly-once contract: client idempotency keys, durable operations claimed with fencing tokens, and credit reservations that commit on success and refund on failure, so retried, concurrent, or crashed requests never double-charge a user. I also built a provider-agnostic LLM layer over OpenAI, DeepSeek, and OpenRouter with per-plan authorization, prompt-injection containment, schema-validated JSON output, and a 20+ suite security test harness. It runs in production on AWS EC2 behind nginx and Cloudflare.
JurisAI
https://www.jurisaiapp.com.brPensador
https://literaturaviva.duckdns.org/Education
Bachelor's Degree in Computer Science
Universidade Presbiteriana Mackenzie - São Paulo, Brazil
Bachelor's Degree in Physics
Federal University of Ceará - Fortaleza, Ceara, Brazil
Certifications
Machine Learning
Itaú Unibanco
Machine Learning
Alura
Skills
Libraries/APIs
React, Pandas, Scikit-learn, Stripe API, API Development, Node.js, REST APIs, Vue 3, Claude API, Google Drive API, NumPy, TensorFlow, Matplotlib, Keras, XGBoost, Stripe
Tools
Codex, Claude Code, Claude, Git, GitHub, Seaborn, Stream Chat
Languages
Python, Python 3, JavaScript, HTML, CSS, SQL, C++, TypeScript
Frameworks
Selenium, Scrapy, Next.js, Tailwind CSS, Flask, React Native
Paradigms
Best Practices, ETL, Automation, Maintainability, Model Context Protocol (MCP), Agile Software Development, Scrum, Database Development, Management
Platforms
Telegram Bot Platform, Amazon EC2, Amazon Web Services (AWS), Vercel, Google Cloud Platform (GCP), Linux, Jupyter Notebook
Storage
Data Pipelines, Data Integration, JSON, Google Cloud, Amazon S3 (AWS S3), PostgreSQL
Other
RAG Architecture, APIs, Problem Solving, Mathematics, Physics, Supabase, FAISS, Agentic AI, Back-end, Full-stack Development, Artificial Intelligence (AI), AI Agents, Generative Artificial Intelligence (GenAI), Agentic RAG Systems, Retrieval-augmented Generation (RAG), Web Scraping, AI Tools, Product Owner, Software Architecture, Agentic Coding, AI Agent Orchestration, Agentic AI Systems, Agentic Workflow Design, System Architecture, Solution Architecture, Data-driven Dashboards, AI Chatbots, Data Engineering, Chatbots, Data Architecture, LLM Agents, Performance, Pipelines, Scalability, Technical Leadership, OpenAI, Workflow Automation & System Integration, CI/CD Pipelines, ELT, LangChain, LLM Integration, Software Development, Product Planning, Marketing Strategy, Integration, University Teaching, Smart Solutions, Research, Team Leadership, FastAPI, Multi-agent Systems, HNSW, Semantic Chunking, K-means Clustering, K-nearest Neighbors (KNN), Machine Learning, ETL for AI, ETL Pipelines, Large Language Models (LLMs), Full-stack, CTO, Third-party APIs, Workflow Automation, Cloud, Product Development, PDF, Open-source LLMs, Document Processing, API Integration, Architecture, SQL Server, Software Engineering, Cloud Architecture, Low Code, Scraping, AI Integration, AI/ML Workloads, Vector Databases, Minimum Viable Product (MVP), Data Science, Scalable Vector Databases, Product Management, Data Security, SaaS, Web Development, Startups, Cloud Computing, Gemini, Dashboards, Front-end, Auditing, Audits, Reliability, Security, Data Modeling, Prompt Engineering, Debugging, Statistics, Amazon RDS, Training, OpenAI SDK, Decision Trees, TSNE, Umap, Random Forests, SVC, Natural Language Processing (NLP), Linear Regression, Bayesian Classifiers, Classifier Ensembles, Support Vector Machines (SVM), Neural Networks, Cursor AI, RAG Pipelines, Law, Business Law, Qwen, AI Product Management, Legal Technology (Legaltech), SaaS Product Management, Semantic Search, Responsive UI, Real-time Streaming, Web App Development, Background Jobs, Function Calling, Payments, Conversational AI, Customer Support, User Profiles, User Analysis, Orchestration, Sales Funnel, CRM, Supabase Auth, Predictive Learning, Clustering, Data Analysis, Data Analytics
How to Work with Toptal
Toptal matches you directly with global industry experts from our network in hours—not weeks or months.
Share your needs
Choose your talent
Start your risk-free talent trial
Top talent is in high demand.
Start hiring