An AI career assistant that scores and rewrites resumes with retrieval-augmented generation and generates full learning courses in the background.
Generic AI resume tools give vague advice because they know nothing about the reader's actual resume or the role. I wanted feedback grounded in the user's own content, and course generation that does not block the web request while an LLM writes dozens of lessons.
Resumes are parsed, embedded, and stored in pgvector. When a user asks for a review, the most relevant chunks are retrieved by cosine distance and passed to Gemini, which returns structured scores and rewrites. Course generation is pushed onto RabbitMQ queues and handled by separate workers for subtopics, content, and bulk jobs.
Embedder
vector encode
RETRIEVAL
pgvector cosine
Gemini structured output
PIPELINE_REPORT
Only the closest resume chunks go into the prompt, which keeps answers specific to the user and keeps token usage bounded.
On startup each worker resets items left in a generating state by a crashed process, so a restart picks up where the last one stopped.
Messages that fail are acknowledged and recorded instead of requeued endlessly, so one bad job cannot block the queue.
I'm looking for backend engineering roles in distributed systems, high-throughput pipelines, or security infrastructure, remote or with relocation. I'm also happy to talk about any of the projects here.
Direct email
rahul0singh003@gmail.comI reply personally, usually within 24 hours.
