Diogo Magalhães Machado

Pipefy's AI Platform runs LLM-based agents (Assistant, Agent, Builder, Analytics) for enterprise customers, including major banks, telcos and global retailers, in single-tenant deployments.

• Joined through the Automations team, working on the core product (Ruby on Rails) and Pipefy's self-hosted iPaaS based on Activepieces (Node.js/TypeScript), before moving to the AI Platform.
• Core contributor to the platform's RAG (Retrieval-Augmented Generation) service: data-source ingestion, embedding and retrieval on pgvector/PostgreSQL, removing the platform's dependency on Pinecone.
• Worked on the production migration from Pinecone to pgvector with a dual-write, dual-read, cutover strategy gated by feature flags, with rollback per tenant and per agent.
• Implemented similarity search and contextual compression retrieval endpoints, backed by asynchronous processing on Celery and Redis Sentinel.
• Deployed the service on AWS, Azure and OCI with Kubernetes/Helm and GitLab CI, including E2E stages running against staging and production.

Python, FastAPI, Celery, Redis Sentinel, PostgreSQL, pgvector, LangChain, LangGraph, OpenAI, Vertex AI, AWS Bedrock, Kafka, Docker, Kubernetes, Helm, GitLab CI, Ruby, Node.js

Risk Management Platform: data engineering behind fraud prevention.

• Developed extraction, loading, transformation and cataloging of data within the warehouse, feeding machine learning models for fraud prevention, dashboards and analysis.
• Built continuous data quality assertion routines for integrity and reliability.
• Engineered data models supporting ML training, monitoring, business intelligence and risk operations.
• Designed and spearheaded Amora, an open source data build tool: data schemas as Python type annotations (PEP 484) and transformations in SQLAlchemy, compiled to warehouse jobs.

Python, SQL, Airflow, BigQuery, dbt, GCP, Feast, Dash, Plotly

Messaging

• Hands-on technical leadership of OLX's chat platform: hundreds of thousands of users connected at all times, 10M+ messages per day, on Elixir/Erlang with Python subsystems.
• Quality and observability work leading to 99.9% uptime and fewer support tickets; messaging flow optimizations cut operation cost while traffic grew 70% year over year.
• Designed image sending in chat with gradual geographic rollout: Elixir/Phoenix media service, CloudFront CDN, automatic content pre-moderation.

Location and Accounts

• Migrated a legacy C localization service used by ad insertion and search indexing to Python AsyncIO, Java Vert.x and Elasticsearch, cutting cost and stabilizing releases.
• Investigated and fixed an iOS memory leak causing out-of-memory crashes since 2013.

Elixir, Erlang, Python, React, React Native, RabbitMQ, Elasticsearch, Redis, AWS

• Led development of ad-delivery APIs serving americanas.com, submarino.com and shoptime.com, with RabbitMQ messaging and Elasticsearch search and recommendation.
• Built data-driven B2B products for marketplace sellers: assortment recommendation from product visualization data and price competitiveness analysis.
• Authored aiologger, an asynchronous, non-blocking logger for Python and asyncio, and contributed to async-worker, an asyncio microframework for AMQP and HTTP consumers, both open source and widely used in production.
• Monitoring and observability of the production environment; hiring and selection of developers.

Python, Node.js, RabbitMQ, Elasticsearch, Docker, Grafana

Price intelligence for Brazilian e-commerce.

• Designed the service keeping millions of marketplace product records up to date: concurrent Python AsyncIO processing hundreds of thousands of messages per minute over RabbitMQ.
• Built the Price Simulator: buybox analysis with win probability, aioHTTP backend serving concurrent users with no cache layer.
• Designed the pipeline integrating tens of thousands of products per minute into B2W's stores.
• Trained the team in asynchronous programming; modeled MongoDB read-optimized structures.

Python, AsyncIO, aioHTTP, RabbitMQ, MongoDB, Django, React, Docker

• Designed and deployed the university web portal: Plone CMS, Varnish caching, Pound load balancing.
• Full-stack process automation web apps in Python (web2py, Flask, SQLAlchemy), including a RESTful API used as an open data provider.
• iOS apps for internal events with QR code reading and LDAP auth; a MeteorJS realtime workflow-mapping editor; network administration and monitoring tooling.

Python, web2py, Flask, SQLAlchemy, Plone, MeteorJS, MongoDB, iOS

• Repairs and preventive maintenance of radars and radio communication devices on the Air Force's transport aircraft.