Car Advisor - Recomendador de Veículos com Claude
aman3096/car-advisor
Recomendador inteligente de carros com IA conversacional e streaming
Ver no GitHubPlugin full-stack que aconselha consumidores sobre a compra de carros usando Claude via Amazon Bedrock. Integra um backend Express com API de streaming e frontend React que coleta preferências do usuário (orçamento, ano, marca) para gerar recomendações personalizadas de um catálogo JSON.
Instalação
npx convez-ai aman3096/car-advisor1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Concessionárias e plataformas de venda de veículos consultivas
- Chatbot de atendimento para orientar clientes na escolha de carros
- Busca inteligente e recomendação de modelos baseada em perfil do comprador
Veredito

Solução prática que demonstra bem a integração do Claude via Bedrock em um caso de uso real (recomendação de produtos). O código é bem estruturado com streaming nativo, rate limiting e separação clara entre camadas, embora reconheça propositalmente limitações como a ausência de RAG e auth em favor do escopo MVP.
README
A tradução para o português deste README ainda está na fila.
car-advisor
A full-stack solution advising customers about the cars using Amazon Bedrock's Claude model, given the JSON of the available cars
Screenshots
Approach
Mock database and use the JSON (skipping the real DB work) There will be a simple express server that will contain api for calling Claude's model Initially, there will be a client that will contain the set of questions asking the users about their budget, year of make, and preferred brand/model Use the questions to give it to the Anthropics' Claude Sonnect, along with the JSON generated, to get our top 3 recommendations from the dataset provided Next would be sending the whole JSON to the customer, and the customer can choose along and may be dig in further with the chat involved
What did you build and why? What did you deliberately cut?
- What did I build?
- Bedrock's streaming API for streaming the responses to the customer's queries
- API Layer using Express
- React application showcasing the responses in a streaming way
- Rate Limiting added What was cut deliberately?
- Building the RAG pipeline was one thing to be deducted due to mock data; in prod scenarios, this should be taken into consideration
- Building real databases and Vector DBs and embedding since Claude's Sonnet model is already doing the work
What’s your tech stack and why did you pick it?
- Amazon Bedrock
- Claude's Sonnet model v4
- Express (API Service layer)
- React (Client)
What did you delegate to AI tools vs. do manually? Where did the tools help most?
- Creation of repetitive data and codes, which can be built faster by Claude rather than a human writing the code
- CSS and UI changes
- Manual work was on the orchestration and making sure that things worked out
- Taking care of the coding standards
- Tools that helped here were Claude Code, Copilot for scaffolding and error handling scenarios, and delivering faster code.
If you had another 4 hours, what would you add?
- Docker Addition could be considered for the deployment part, used to render because of the time constraints and it being a free one
- Auth Layer and Authorization Layer
- RAG pipeline (It would refer to the Internal data and will not expose the data to the training model (Anthropic))
- Further improvements in the codebase.