A conversational job search prototype

Job Manager is a proof of concept for a modern career-services platform. I built it with another engineer to explore how natural-language preferences could help job seekers find relevant opportunities.

The central interaction is a chat with an agent: describe your background and the kind of work you want, then let the system refine the listings around those signals. The demo below follows a JavaScript-focused, remote-work search.

Architecture and technology

The application combines a Next.js frontend with a NestJS backend on serverless AWS infrastructure: Lambda, S3, and CloudFront. The matching engine uses LangChain RAG, OpenAI GPT-4o-mini, and OpenSearch vector search for semantic job discovery. Infrastructure is defined with OpenTofu across development, production, and administrative AWS accounts, with GitLab CI/CD using OIDC authentication.

What I learned

The project made the value of an LLM-driven interface concrete. Job search is unusually personal: each opportunity and each person has different context, so fixed filters can only approximate what someone means. Language creates room for a more direct conversation about the outcome.

It also reinforced where I do my best work: owning the path from the backend to the customer problem, setting the priorities, and shaping the process that delivers the product.