Search

Search pages and journal entries…

Search

Search pages and journal entries…

Leonardo Ferreira

Product Engineer · Rio de Janeiro, Brazil

Leonardo Ferreira

I turn user problems into software, from the first question to production.

5+ years across B2B SaaS, backend, data engineering, product leadership, and applied AI.

Explore my work Get in touch

Contact and current work

Senior Software Engineer @Reach.ai

[Email protected]

UTC−3 · Remote

Portuguese (native) · English (fluent)

Social Links

LinkedIn

leonardo-v-ferreira

GitHub

leovferreira13

About

I am a Product Engineer with 5+ years across backend and data engineering, product management, and technical leadership. I connect user and business needs to practical software, taking problems from discovery through technical decisions, implementation, launch, and iteration.

Adaptability is the thread through my work. I enjoy learning an unfamiliar domain, listening to the people who know it, and working through ambiguity to understand what is useful to build. I am comfortable moving between customer conversations, product trade-offs, and the technical details that make a solution work in practice.

I bring strong backend foundations and product judgment to that process, with experience in APIs, data pipelines, integrations, and event-driven systems. More recently, I have been applying that background to AI agents, focusing on how they connect to real workflows, behave reliably, and can be evaluated and improved. I work best close to the problem, with ownership of both the engineering and the experience it creates.

Experience

REACH.ai

Present
  • Engineer Python/FastAPI APIs and event-driven messaging flows for a multi-tenant SaaS platform processing 20M+ SMS and emails monthly, with 99.9% delivery reliability. The system connects AWS SQS, Lambda, API Gateway, Aurora/RDS, MongoDB, Redis, and messaging providers.
  • Improved queue processing and introduced Redis-backed idempotency, reducing duplicate-send risk and supporting 50x peak throughput capacity.
  • Building a multi-turn voice and SMS appointment agent with Python, LangGraph, ReAct-style orchestration, OpenAI tool calling, and RAG over business knowledge. Integrate scheduling APIs for availability, booking, rescheduling, and cancellation.
  • Designed behavioral evaluations, guardrails against unsupported responses, and observability through LangSmith and internal audit logs, so conversation behavior and tool calls can be inspected and improved.
  • Optimized median voice response latency to approximately 2 seconds in internal tests. The agent is in prelaunch testing; customer adoption and operational outcomes have not yet been measured.
  • Python
  • FastAPI
  • AWS Lambda
  • SQS
  • Redis
  • MongoDB
  • LangGraph
  • RAG
  • Evaluations

Guiavet

  • Led five engineers, owning prioritization, technical decisions, and delivery as the B2B veterinary SaaS product matured.
  • Owned technical continuity of the platform while reducing infrastructure and operational costs through automation and lean delivery.
  • Led major improvements to the Health Records module: integrated drug reference data, built a pharmacy dispensing portal with external validation, customizable prescription templates (faster daily record creation), and customizable document workflows.
  • Re-architected multi-page PDF generation and asynchronous processing at Guiavet, reducing generation time from approximately 90 to 15 seconds (about 83%), improving a document workflow used by veterinary teams.
  • Delivered end-to-end product changes across backend (Python/Django), frontend (Vue.js), and AWS operations while coordinating with senior engineers under Agile practices.
  • Python
  • Django
  • PostgreSQL
  • AWS
  • Vue.js
  • SaaS
  • PDF generation
  • Agile
  • Product management
  • Roadmapping
  • UX
  • B2B
  • SEO
  • Agile
  • Python
  • Django
  • PostgreSQL
  • AWS
  • SQS
  • ETL
  • SQL
  • nginx
  • DDD

QDOIS - Data Intelligence

  • Azure Data Factory
  • Azure Databricks
  • PySpark
  • ETL
  • Azure DevOps
  • Git

Selected work

All projects
Football, Visualized: interactive pitch, expected goals estimate, and model features

Public project · Oct 2026

Football, Visualized

An interactive football analytics app that makes expected goals tangible: move a shot, inspect its features, and explore the model behind the estimate.

The interactive pitch lets visitors change shot conditions and inspect eight model features alongside an expected-goals estimate. I connected a reproducible StatsBomb Open Data pipeline to XGBoost training and evaluation, then brought inference into the browser with a Next.js/React interface.

The playground is available in five languages and includes methodology and evaluation results. It brings my data, applied ML, backend-to-interface integration, and product work into one public project: visitors can explore what the model does, while the repository exposes how the result was built.

  • Next.js
  • React
  • TypeScript
  • Python
  • XGBoost
  • StatsBomb Open Data
Open live app View source Journal

Capabilities & tools

Product & delivery

I connect interviews and usage data to workflow design, prioritization, and technical scope. At Guiavet, this included adapting a veterinary product to different B2B segments and leading five engineers through delivery.

Backend & cloud

Python is my main backend language. I use FastAPI for APIs, Django for domain-heavy products, PostgreSQL and MongoDB for persistence, and Redis and AWS queues for asynchronous processing and idempotency.

Data engineering

I build ETL pipelines and data models that connect operational systems to reporting and ML. My consulting work used Azure Data Factory, Databricks, and PySpark to integrate databases, APIs, and SFTP sources.

Applied AI

I use LangGraph to orchestrate multi-turn agents, OpenAI tool calling to act through APIs, and RAG to retrieve business knowledge. Behavioral evaluations, guardrails, LangSmith traces, and audit logs help inspect behavior and failures.

Frontend

I work across the interface when a product needs it: Vue.js in veterinary SaaS, and React, Next.js, and TypeScript in Football, Visualized. The goal is to make backend and model behavior understandable to the user.

Education

  • Bachelor's degree, Control and Automation Engineering.
  • Strong foundation in mathematics, programming, and systems engineering (control systems, automation, and software development).
  • Control systems
  • Automation
  • Software engineering
  • Mathematics
  • Mechatronics
  • Robotics
  • Automation

Publications(1)

Explores algorithmic recourse in mental healthcare contexts, including a classification framework with observed and latent variables.

Journal

No published entries yet. This is where I will share project decisions, release notes, experiments, and the evidence behind them.

Explore the journal