15 years shipping production backend systems in Go and Node.js. Last 18 months going deep on applied AI: building products end-to-end from Kubernetes infra up to RAG pipelines and streaming LLM SDK internals.
I work as an independent contractor through MVN Development LTDA (Brazil) for remote teams worldwide. USD only, full overlap with US Eastern.
@langchain/openai (langchain-ai/langchainjs), shipped in v1.4.6
Streaming structured outputs against the OpenAI Responses API were dying mid-flight when models like gpt-5-mini emitted trailing non-whitespace after the JSON object. The bare JSON.parse in the Responses delta converter would throw a SyntaxError that killed the entire stream. I wrapped the parse in try/catch and let the existing withStructuredOutput pipeline handle the typed failure: includeRaw:true returns { raw, parsed: null } via withFallbacks, includeRaw:false throws a typed OutputParserException the caller can catch and retry. 2 unit tests + 3 E2E tests, approved by christian-bromann and bundled in the same patch release as his own maintainer fix.
Playsonora, an AI-powered music school. Solo build.
Backend, data layer, RAG pipeline, and human-in-the-loop teacher review. All me. Engineering notes on the blog.
- Production-grade Go and Node.js microservices: NestJS, RabbitMQ, event-driven, sub-200ms p95
- LLM systems in production: RAG pipelines, multi-provider routing (OpenAI / Gemini / OpenRouter), structured outputs, LLM-as-judge evaluation, human-in-the-loop review loops
- Streaming SDK internals: hands-on with the OpenAI Responses API, LangChain.js
withStructuredOutput, error-propagation semantics of streaming LLM calls - Cloud and infra: Kubernetes/GKE on GCP, Terraform, OpenTelemetry + Prometheus, multi-cloud (GCP / AWS / fly.io)
- End-to-end ownership: from cluster up to product features. One hire instead of three.
| Where | What | Numbers |
|---|---|---|
| Digai (current, since May 2025) | Senior Backend on AI-WhatsApp interview product: candidate-journey state machine, LLM-as-judge prompt simulator over 8 red-team scenarios, multi-provider LLM routing | Multi-tenant SaaS, async event-driven (SQS / EventBridge), DynamoDB high-volume conversation log |
| Air Apps | OpenAI integrations + ETL/ELT rewrite | +25% engagement, -50% incident response, -40% data processing time, <200ms p95 |
| Eleven Dragons | Tech Lead Backend, NestJS + Go microservices on fly.io / AWS / GCP | OTel + Prometheus instrumented, mentored junior |
| Mobiplus / Afya | CliqueRX, digital prescription system for Brazilian doctors | Shipped to millions of users |
Full track record: LinkedIn
Go TypeScript Node.js NestJS PostgreSQL Redis DynamoDB RabbitMQ SQS EventBridge Kubernetes GKE Terraform Docker OpenTelemetry Prometheus OpenAI LangChain.js Structured Outputs RAG
Senior or Staff Backend roles, 100% remote, USD, where AI is a core product capability or backend reliability at scale is the core challenge.
Talk to me: viniciusmvn@pm.me · LinkedIn DM