Skip to content

Available · remote · B2B contractor, international contract or full-time (CLT, Brazil)

AI Engineer.

I build LLM, agent and RAG systems and ship them to production, from architecture to infrastructure.

I deliver complete systems with AI built in (SaaS for clinics, fintech, mobility, clinical support), from the app to the backend, with an agent-first workflow.

Book a callView casesDownload CV (PDF)
Everton Resende

Fact sheet

Focus
LLMs, agents and RAG in production
Career
7+ years in software development
Stack
TypeScript · Python · Cloudflare · AWS · Postgres
Location
Pará de Minas, MG · remote (UTC−3)
Status
Available · B2B contractor, international contract or full-time (CLT, Brazil)
  • Products

    5products in production

  • TazzIn production

    AIAI agent in customer support, under human review

  • MaximusIn production

    RAGClinical RAG chat used by anesthesiologists

01 / What I deliver

From architecture to production.

1.1

I architect agents.

Tool calling, layered guardrails, LLM‑as‑judge, evals and model fallback, in TypeScript. In Python, agent teams with triage and handoff (Agno).

ProofTazz agent · Python agents (Agno) at Coppay

1.2

I put RAG into production.

Native vector search in production, replacing the classifier-based manual RAG I built first. And I know when RAG is not worth it.

ProofMaximus · the no-RAG decision at Tazz

1.3

I build and run the infra.

Cloudflare (Workers, Durable Objects, Queues), Postgres with RLS, Docker and CI. On AWS: EC2, ELB, Lambda, SQS, S3, Aurora and Rekognition.

ProofTazz migration · Coppay · Dental Smart · AWS in earlier projects

02 / Cases

Shipped products. Not demos.

Mobility · 2026–present

Tazz.

View case
Fig. 01Architecture
  1. Expo app
  2. Workers
  3. Postgres

New app from scratch · coding agents in parallel

Problem
The ride-hailing app’s backend hit GCP’s fixed 20 vCPU quota in São Paulo, and the Flutter app talked straight to Firebase.
What I did
I architected and built the new Expo app and its new Workers from scratch, with coding agents working in parallel (pre-launch). Before that, I ported 100+ Cloud Functions to 8 Cloudflare Workers in ~47.5 h. I also built the support agent and took PIX with Asaas to production.
Stack
TypeScript · Expo · Cloudflare Workers · Durable Objects · Queues · Postgres
AI
In productionSupport agent with tool calling, guardrails, LLM‑as‑judge and evals, with a person reviewing every reply.
Result
PIX in production since Sep 2026
SaaS for clinics · 2025–present

Dental Smart.

View case
Fig. 02Transcription and summary
  1. Appointment
  2. Whispertranscription
  3. Llamasummary

Multi-tenant · RLS isolation

Problem
Dental clinics needed finance, NFS-e invoices and WhatsApp messages in one system, with each clinic’s data isolated.
What I did
I’ve been building the multi-tenant SaaS: AI appointment transcription, RLS isolation, scheduled WhatsApp messages, finance and NFS-e invoices.
Stack
Next.js · React · Supabase · Postgres RLS · Edge Functions · Groq
AI
In productionAppointment transcription and summary (Whisper + Llama).DecisionBank reconciliation is rule-based, no AI.
Result
~20 clinics · 0 churn
Healthcare · 2026

Maximus.

Fig. 03RAG
  1. Questiontext, image or voice
  2. Responses APIgpt-5.4 · streaming
  3. file_searchOpenAI vector store

Before: classifier + manual search. Now: native file_search

Problem
Anesthesiologists needed fast clinical answers (doses, drugs, protocols) backed by a reference knowledge base, plus practice questions for Brazil’s anesthesiology board exam (TSA).
What I did
I built the app in Expo and the API in Hono on Cloudflare Workers, with Supabase and RLS. The chat uses the Responses API with file_search over the reference knowledge base and code interpreter for charts, with streamed answers.
Stack
TypeScript · Expo · Hono · Workers · Supabase · OpenAI Responses API
AI
In productionClinical chat with RAG over the reference knowledge base, with questions by text, image or voice (gpt-4o-transcribe).In productionBoard-exam-style (TSA) questions generated with structured output and validation, from the vector knowledge base.DecisionI built manual RAG (a gpt-5-nano classifier + vector search), then switched to native file_search: the classifier was blocking legitimate searches.
Result
Live, used by anesthesiologists
Fintech · 2025–2026

Coppay.

Fig. 04Architecture
  1. AppExpo
  2. BFFHono · Workers
  3. BaaSCronos → Asaas

Back office in Next.js · RBAC · audit log

Problem
A fintech needed a digital account, credit and customer support running on a BaaS, from the app to the back office.
What I did
I built the Expo app, the Hono BFF on Cloudflare Workers (Cronos, Asaas and HBI), the Next.js back office with RBAC and an audit log, and the AI features: email triage, LLM-assisted credit reviews and the WhatsApp agent. In Sep 2026, I built a new BFF on the Asaas BaaS, with one subaccount per customer, running side by side with the legacy one.
Stack
TypeScript · Expo · Hono · Cloudflare Workers · Durable Objects · Postgres · Python · FastAPI
AI
In productionEmail triage (Workers AI), used by the team, and LLM-assisted credit reviews (gpt-oss-120b), used by the analysts.Offline todayWhatsApp agent in Python (Agno + FastAPI) with triage to specialists, read-only tools and human handoff. It served real customers until Sep 2026.
Result
App in the stores with real account holders
SaaS for photographers · 2026–present

Davions.

Fig. 05Architecture
  1. AppReact
  2. 10 Workers
  3. Supabase (RLS) · R2

Brazilian escrow on Asaas · SQL ledger

Problem
A gallery SaaS for photographers, first built in Lovable, needed to be re-hosted and refactored onto its own architecture.
What I did
I migrated and refactored the app onto 10 Cloudflare Workers with Supabase and RLS, and architected the escrow with Asaas: an SQL ledger and PIX payouts.
Stack
TypeScript · React · Cloudflare Workers · R2 · Supabase RLS · Asaas · Stripe Connect
AI
In productionBlog post, caption and theme generation with Claude, through Cloudflare AI Gateway.
Result
Delivered in Oct 2026 · escrow in dark launch

03 / How I work

Agents write the code. I own it.

I work with agents the way I’d lead a team. I set the context, the architecture and the acceptance criteria. The agents write most of the code. I read it, debug it, require a green test suite before every commit and decide what goes to production.

Pipeline

  1. 01SpecDefinition of done, as verifiable criteria.
  2. 02PlanSmall tasks, files and order.
  3. 03Parallel subagentsOne workstream per worktree.
  4. 04ReviewCode review and a second opinion from another model.
  5. 05TestsGreen suite before every commit.
  6. 06PreviewA preview version before deploying.
  7. 07ProductionDeploy with a documented rollback.
3.1

I decide

  • Architecture and trade-offs
  • What goes in and what stays out
  • Acceptance criteria
  • Where not to use AI
  • What goes to production
3.2

I delegate to agents

  • Writing the code
  • Bulk sweeps and migrations
  • First draft of the tests
  • Researching docs and repositories
Quality
CLAUDE.md per repository · acceptance criteria · review · tests as a commit gate
In practice
At Tazz, coordinator agents ran a backlog of ~200 workstreams for the new app in parallel sessions, one worktree each, reviewed by another model.
This site
It was rebuilt this way. The spec and the plan are in docs/ on GitHub.See the spec

04 / Experience

Experience.

Oct 2025 – Oct 2026
AI EngineerCoppayFull-time (CLT) · remote
Nov 2022 – Oct 2025
Software DeveloperSoftplanFull-time (CLT) · remote
Nov 2018 – Oct 2021
Software DeveloperFAPAMFull-time (CLT) · hybrid
4.1

Education

  • Information SystemsFAPAM · Nov 2018 – Dec 2020 · not completed
4.2

Open source

05 / Contact

Let’s talk.

Available · remote · B2B contractor, international contract or full-time (CLT, Brazil)

Book a callDownload CV (PDF)

Contact details

Email
everton.augustor@gmail.com
LinkedIn
linkedin.com/in/evertonresende
GitHub
github.com/evertonresende
WhatsApp
+55 37 99913-3556