Victor LaperutaLead full-stack engineer · AI agents in production

Plain version
LinkedIn
Medium

Contact

Victor Laperuta

Lead full-stack engineer · São Paulo, Brazil

Open to lead and senior full-stack roles, remote.

victorlaperutademoura@gmail.com

LinkedIn Medium

NowSummarySummarySão Paulo, BrazilLead full-stack engineerWho I am, what I read, and where this site was built.TypeScriptNestJSReactPostgreSQLAWSLangGraph
2020Marketplace catalogue2020 · CatalogueDolado, São PauloFull-stack engineer, first engineering hireA free catalogue for neighbourhood shops during the pandemic: the site, the backend, the app.Next.jsReactReact NativeExpressMySQL
2021–2023B2B marketplace & integrationsB2BDolado, São PauloFull-stack engineer, de facto tech leadA B2B marketplace for small retailers: the storefront, the typed monolith, the credit analysis.TypeScriptExpressShopifyKondutoPythonNestJS
2023–2025Legacy commerce platformPlatformDolado, São PauloBackend engineer, largely aloneAn acquired dropshipping platform, kept running largely alone; its failures found and fixed.AWS LambdaAmazon SQSMySQLCronicleTerraform
Jun 2025–Jan 2026AI agent platformAI agentsDolado, São PauloHead of TechnologyA team of five, an agent that failed in production, and the redesign that took it to weekly use.LangGraphClaudeOpenAIModel Context ProtocolNestJSLangSmith
2026A product of my ownMy productSão PauloFounder, sole engineerAn audio product built end to end in two months, tested, and paused on purpose. Then reading and writing.NestJSPostgreSQLPythonLiteLLMExpoNext.js

Now

Summary

São Paulo, Brazil

Lead full-stack engineer

Who I am, what I read, and where this site was built.

TypeScriptNestJSReactPostgreSQLawsAWSLangGraph

All projects● Now · Summary 2020 · Marketplace catalogue 2021–2023 · B2B marketplace & integrations 2023–2025 · Legacy commerce platform Jun 2025–Jan 2026 · AI agent platform 2026 · A product of my own

Victor Laperuta

Lead full-stack engineer · São Paulo, Brazil

I build products from the first commit, keep them running in production, and took a team to LLM agents in production.

Six years at Dolado, from first engineering hire to Head of Technology, through four pivots.

Open to lead and senior full-stack roles, remote.

victorlaperutademoura@gmail.com Read the one-minute version

What I build with

TypeScript and Node, NestJS on the back, React and React Native in front, PostgreSQL and MySQL, AWS with Terraform, LangGraph and MCP when there is an agent in it.

Education: B.S. Computer Engineering, Insper, São Paulo, 2016 to 2021. Honours for the capstone: a mobile app for visualising an investment portfolio.

Courses: Tech Ethics: Critical Thinking in the Age of Apps, Algorithms, and AI (HarvardX, completed 2026); Digital Wellbeing (University of York, in progress).

Languages: Portuguese, native; English, fluent.

Tools

TypeScriptNode.jsNestJSReactRNReact NativePostgreSQLawsAWSTerraformLangGraphMCPModel Context Protocol

The games on the desk

LittleBigPlanet (Media Molecule, 2008, PlayStation 3). Sackboy, a world of cloth, cardboard and stitches, and a level editor that let anyone build. The look of this site comes from it.

LittleBigPlanet 2 (Media Molecule, 2011, PlayStation 3). The sequel: the same craft world, with a creator kit deep enough to build whole games inside it.

Uncharted: Drake’s Fortune (Naughty Dog, 2007, PlayStation 3). Nathan Drake’s first hunt, a lost island and a playable adventure film.

Uncharted 2: Among Thieves (Naughty Dog, 2009, PlayStation 3). The train, the Himalayas, Shambhala: the one everyone remembers.

Need for Speed: Underground 2 (EA Black Box, 2004, PlayStation 2). Night-time street racing in a city you could drive around, and the tuning that made each car yours.

LittleBigPlanet ↗LittleBigPlanet 2 ↗Uncharted: Drake’s Fortune ↗Uncharted 2: Among Thieves ↗Need for Speed: Underground 2 ↗

The app, and what was under it

Screens: home, regions, location, search, shop, product, checkout, payment method, card, login, signup.

React Native, Redux, OneSignal for push, geolocation for the shops near you.

Backend: Express on MySQL. Search with Algolia. Payments with Asaas, card and bank slip. SMS with Twilio. Photos on S3. Delivery through a courier API.

The link a shop shared on WhatsApp opened the same catalogue on the web.

Tools

RNReact NativeExpressMySQLAlAlgoliaAsAsaasTwTwilioOSOneSignalS3Amazon S3WhatsApp

The website

Next.js with React and Material UI, deployed on Vercel.

Pages: search, shop signup, cart, checkout, account, help; the merchant side with my sales and product registration.

An end-to-end suite with Nightwatch, GitHub Actions, and Docker gated the deploys in 2020. It rotted later, which taught me what a test suite costs.

Tools

Next.jsReactMUIMaterial UIVercelGitHub ActionsDocker

The storefront

A headless Shopify shop: Next.js on the Storefront API with GraphQL and Apollo, the custom React front the merchants used to order.

The Shopify service received the webhooks and kept products, stock and orders in step with the ERP; it moved to NestJS in March 2023.

Tools

ShopifyNext.jsGraphQLApollo GraphQLNestJS

07:30 · invoices in batch, hundreds of orders

a mysql connection per php file · the rds limit blew · logistics froze

a seller’s refresh token · every two months

whoever forgot · orders stopped · nobody knew

ShopeeRefreshToken · logistics_channel_list

came back as objects · the job crashed until the fix

catch (Throwable $e) {}

the error vanished · the job died without a log

an aws key, exposed by an insecure php app

a rogue instance sending mail · killed, traced in cloudtrail, rotated

Three things I would tell another CTO

Start flexible for discovery, add structure for reliability. The monolithic agent found the real needs; the specialised nodes made them dependable.

Adoption is an organisational problem. What won it: the agent ran in parallel with the spreadsheets for weeks, so the analysts compared its results with their own and trusted it on their own terms. The reframing that held: the intelligence team stays the owner of the process and its quality; their role moves from manual execution to insight and the client; the agent is their tool.

Track usage to discover value.

Tools

LangGraphClaudeoaiOpenAI

agent · step 14 · sql

SELECT data, total FROM orders … column ‘data’ does not exist

What people used it for

We planned one use: the default market report, a whole subcategory over a period.

By tracking usage, four more appeared: deep-dives on a few sections with the terms to ignore; ad-hoc questions (“the top brands by revenue”) without a full report; walking the category tree for other research; extracting prices from specific listings for competitive analysis.

From the chat: a report, a presentation, charts asked for in plain language, all generated from the same analysis.

The agent went from a report generator to a research copilot, because it stayed conversational and we watched how it was used.

Tools

ClaudeoaiOpenAIPrPresenton

The app and the architecture

Human-in-the-loop pauses and visible tool calls while an episode is built.

The player shows the transcript as it is read.

Four repositories: the API, the app, the site, the LLM and speech service. About a thousand commits in about two months.

API: NestJS, hexagonal (interfaces, domain, infrastructure), PostgreSQL, TypeORM, Redis, BullMQ, SSE, JWT, an MCP server.

Speech and models: LiteLLM in front of Anthropic, OpenAI, DeepSeek and Gemini, with Groq and a local Ollama as fallbacks; Kokoro and XTTS for speech; Whisper for the voice-set weekly goals; Docker Compose for all of it.

App: Expo, React Native, Tamagui, Zustand, Reanimated. Site: Next.js 15, React 19, Chakra UI v3, Playwright.

Quality: promptfoo evals gate every prompt change.

Every repository blocks a commit that fails the architecture read, the tests, or lint.

Tools

ExpoRNReact NativeTgTamaguiNestJSPostgreSQLMQBullMQRedisLLLiteLLMOllamaTTSKokoro TTSSTTWhisperXDockerNext.jsChakra UIPwPlaywrightpfpromptfooLfLangfuse

The app and the architecture

Human-in-the-loop pauses and visible tool calls while an episode is built.

The player shows the transcript as it is read.

Four repositories: the API, the app, the site, the LLM and speech service. About a thousand commits in about two months.

API: NestJS, hexagonal (interfaces, domain, infrastructure), PostgreSQL, TypeORM, Redis, BullMQ, SSE, JWT, an MCP server.

App: Expo, React Native, Tamagui, Zustand, Reanimated. Site: Next.js 15, React 19, Chakra UI v3, Playwright.

Quality: promptfoo evals gate every prompt change.

Every repository blocks a commit that fails the architecture read, the tests, or lint.

Tools

ExpoRNReact NativeTgTamaguiNestJSPostgreSQLMQBullMQRedisLLLiteLLMOllamaTTSKokoro TTSSTTWhisperXDockerNext.jsChakra UIPwPlaywrightpfpromptfooLfLangfuse

What I read and write now

Reading: Children’s digital protection in Brazil. Law 15.211 of 2025, the ECA Digital, in force since 17 March 2026: the first Brazilian law to put rules and penalties directly on digital platforms where children and adolescents may be. I have been reading the law and the decree and mapping what each kind of product owes.

Reading: Attention, and what software does to it. The 2026 shelf: twenty-odd books on attention, technology, and the mind, read this year. Two courses this year: Tech Ethics (HarvardX, completed) and Digital Wellbeing (University of York, in progress).

Writing: A Janela Imprevisível (2026, in Portuguese). Why the uncertain latency and quality of AI tools work on developers like a slot machine.

Writing: A nova economia do software (2026, in Portuguese). Principles for building software when production is cheap and judgment is the bottleneck; an eleven-slide carousel on LinkedIn.

Writing: End-to-end tests for Next.js with Nightwatch, GitHub Actions, and Docker (2020, in Portuguese). A suite that gated deploys, and later rotted, which taught me what a test suite costs.

A Janela Imprevisível ↗A nova economia do software ↗End-to-end tests for Next.js with Nightwatch, GitHub Actions, and Docker ↗