Currently

Shipping production LLM systems on top of solid backend foundations.

Event-driven RAG pipelines, MCP integrations, and multi-tenant SaaS backends in production: 1M+ RAG answers served, with infrastructure cost cut by ~50% through orchestration and retrieval re-engineering.

5+ years building Node.js, TypeScript, AWS, React, and applied AI in production. Mechatronics Engineering at USP, AWS Certified.

View work
01 /

About

I'm a Fullstack AI Engineer with 5+ years building production systems where AI is the feature, not the demo. Backend-leaning by instinct, fullstack by delivery.

My work lives at the intersection of distributed backends and applied AI: event-driven architectures on AWS, multi-tenant SaaS, and LLM systems shipped to real users. In my most recent role I worked on a messaging-integration platform (bot and reputation channels routed into chat and helpdesk platforms) that has processed 8.6M+ customer messages across 38 tenants. There, I designed the event-driven RAG pipeline and built the product's first MCP layer.

Background in Mechatronics Engineering at USP, now being deepened by a postgraduate program in Applied AI. I treat AI like any other engineering surface: trade-offs, evaluations, cost ceilings, and observability. I'm pushing toward senior backend / AI engineering, with a focus on international remote roles.

02 /

Selected Work

Event-driven RAG over multi-tenant knowledge bases

Problem
A multi-tenant B2B platform's conversational AI answered from each tenant's own knowledge base, but infrastructure cost was scaling faster than usage as tenants and LLM workflows multiplied.
Approach
Re-engineered the pipeline into decoupled event-driven stages on AWS, restructured retrieval around Typesense, and consolidated orchestration so multi-LLM conversational workflows share a single retrieval layer.
Impact
Cut RAG infrastructure cost by ~50% while serving production multi-tenant traffic: 1.2M+ answers served to date, peaking at 148k in a single month. The pipeline became the foundation for every AI feature shipped after it.
NestJSTypeScriptLangChainOpenAITypesenseAWS

Proprietary production system (NDA) · architecture walkthrough on request

droz-claude: an internal AI engineering workspace

Problem
AI-assisted development was individual and ad-hoc: no shared domain context, no common tooling, results varying widely between developers.
Approach
Designed and shipped a team workspace built on Claude Code: custom MCP servers, per-domain context files, reusable skills, and a code-intelligence graph over the team's main repositories.
Impact
In the 49 days after team-wide adoption: tasks closed per day rose 36%, code shipped per day nearly doubled, average PR size grew 144%, and bugs created per day fell 59%. Five independent signals from Jira, GitHub, and bug tracking moved in the same direction.
Claude CodeMCPTypeScriptGitNexus

Correlation measured over 49 days of team use, not a controlled experiment

Product MCP layer for context-aware AI workflows

Problem
AI features needed structured, governed access to platform capabilities; one-off integrations were starting to multiply across teams.
Approach
Designed and shipped the product's initial MCP layer, exposing platform capabilities as typed, reusable tools with structured access for AI workflows.
Impact
Became the foundation for context-aware AI-assisted workflows and standardized how teams build on the platform's AI surface.
MCPTypeScriptNestJSZod structured outputsOpenAI

Proprietary production system (NDA) · architecture walkthrough on request

Peak-load database contention, from hours to seconds

Problem
A core platform flow saturated MySQL at peak: CPU pinned and request queues backed up for hours.
Approach
Profiled the hot path and introduced token caching with TTL strategies, eliminating redundant queries without changing the public contract.
Impact
Peak request wait time dropped from hours to seconds on production load, with no hardware changes.
Node.jsTypeScriptMySQLAWS

Production incident work (NDA)

03 /

Experience

  1. 2022 2026

    Fullstack Software Engineer · AI Focus·DROZ

    • Owned AI feature delivery end-to-end on a multi-tenant, event-driven B2B SaaS; designed the production RAG pipeline and the initial MCP layer (case studies above).
    • Shipped embedding-based NLP classification (cosine similarity), LLM intent prediction with structured Zod outputs, and image metadata via vision models in production.
    • Designed an internal AI engineering workspace unifying domain context, reusable skills, and MCP tooling across teams.
    • Maintained AWS Lambda workloads and event-driven messaging integrating Zendesk, ReclameAqui, proprietary chat systems, and CS tools.
    NestJSTypeScriptLangChainOpenAITypesenseMCPGraphQLPostgreSQLDynamoDBAWSCDK
  2. Jan 2025 Mar 2025

    Fullstack Software Engineer · Contract · ASAP Codes

    • Fixed-scope contract building backend services in NestJS for a white-label fintech platform.
    • Delivered React / Next.js frontend features wired to MongoDB-backed flows.
    NestJSTypeScriptReactNext.jsMongoDB
  3. 2021 2022

    Junior Fullstack Software Engineer · DROZ

    • Resolved severe MySQL contention and CPU saturation through token caching and TTL strategies. Cut peak request wait times from hours to seconds.
    • Contributed to a low-code cloud platform for external API integrations on AWS using Serverless Framework and clean architecture.
    Node.jsTypeScriptNestJSReactMySQLAWSServerless Framework
04 /

Stack

AI / LLM

  • RAG
  • MCP
  • LangChain
  • OpenAI
  • Gemini
  • Embeddings
  • Typesense
  • Zod structured outputs
  • Vision models

Backend

  • Node.js
  • TypeScript
  • NestJS
  • GraphQL
  • Python

Data

  • PostgreSQL
  • DynamoDB
  • MongoDB
  • MySQL
  • Redis

Cloud / Infra

  • AWS Lambda
  • API Gateway
  • SQS
  • SNS
  • S3
  • EC2
  • CloudFormation
  • CDK
  • Docker
  • Serverless Framework

Frontend

  • React
  • Next.js

Architecture

  • Event-driven
  • Multi-tenant SaaS
  • Distributed systems
  • Serverless

Stack reflects what I currently ship in production, not a wishlist. Comfortable picking the right tool when a project demands it.

05 /

Education

  1. 2026 2027

    Postgraduate Degree · Applied Artificial Intelligence·UNIPDS

    • In progress (Apr 2026 to May 2027): LLMs, RAG, autonomous agents, MCP, and AI-first architectures. Grounding production experience in formal theory.
    LLMsRAGMCPAI agents
  2. 2025 2028

    AWS Certified Cloud Practitioner·Amazon Web Services

    • Issued Oct 2025, valid through 2028, on top of daily production work with Lambda, SQS/SNS, API Gateway, and CDK.
    AWS
  3. 2014 2021

    B.Sc. Mechatronics Engineering · University of São Paulo (USP)

    • Bachelor thesis: Cloud Annotation Library, a platform for sharing academic annotations using MEAN stack and AWS S3.
    • Multidisciplinary engineering foundation: control systems, embedded systems, software, signal processing.
    MEANAWS S3AngularNode.js
06 /

Trusted Network

People I'd hire without thinking twice.

Engineering

EugenioBonadio

Software Engineering Manager

Engineering Manager with a strong technical background. Leads complex projects balancing quality, planning, and individual growth via 1:1s and mentorship.

PT

FrancisTarganski

Software Engineer · Fullstack, Frontend-leaning

Frontend-leaning fullstack engineer, 5+ years across React, Next.js, Node, and conversational AI. Architecture background sharpens UI craft.

ENPT

GabrielBaldino

Software Engineer · Fullstack, Backend-leaning

Backend engineer specialized in Node.js, NestJS, and AWS serverless. 3+ years leading DrozChat, a multi-tenant platform with 100+ enterprise clients.

PT

TiagoVeronesi

Software Engineer · Fullstack, Backend-leaning

Backend engineer with Node.js, NestJS, AWS, and PHP. Built large-scale marketplace and payment integrations across e-commerce and CRM.

PT

Product & Design

MatheusOliveira

Senior Product Manager

Senior Product Manager, 7+ years scaling SaaS across conversational AI, fintech, and enterprise. 0→1 launches, PCI gateways, data-informed decisions.

ENPT

MayanaPirovane

Senior Product Manager

Specialist Product Manager, 6 years across high-scale critical systems. Currently leads OMS at Grupo Boticário, after product leadership at PicPay.

ENPT

ThomasGuimaro

Senior Product Designer

Senior Product Designer, 8+ years of end-to-end product design across EdTech, museums, and AI-powered B2B. MBA in Neuroscience and Consumer Behavior.

PT

ViniciosNogueira

Product Owner

Product-oriented Service Delivery lead, 10+ years coordinating implementation projects end-to-end across support, dev, and customer teams.

ENPT

Let's talk.

Open to international remote Fullstack AI and AI Engineering roles. Email is the fastest channel.

andre.94.cavalcante@gmail.com
LinkedIn

Get in touch

Open to international remote roles in Fullstack AI and AI Engineering. Reach me through any of these.

Email· preferred

andre.94.cavalcante@gmail.com

LinkedIn

in/andre94cavalcante

GitHub

@andre94cavalcante