Applying for developer jobs in Germany means many applications in two languages, each needing the right CV, a cover letter in the correct format and a record of where it went. Generic chatbots don't know your CVs, invent experience and forget everything. Job Application Agent is an AI agent that does this work from the user's own documents.
What it does
- Answers questions from the user's own CVs and cover letters (PDF, DOCX, TXT), with citations to the source
- Finds live job postings from Arbeitnow and Himalayas, and LinkedIn on request
- Picks the best-fitting CV for a job, or tailors a new one as a LaTeX PDF
- Writes cover letters in the job's language (English or German) in the German DIN 5008 format, as Text, Word and PDF
- Tracks every application and remembers the user's preferences across chats
What makes it different
- The agent chooses its own tools. A LangGraph agent decides which of its 9 tools to use and in what order, so one chat covers the whole workflow from finding a job to applying.
- Corrective RAG. After retrieval, an LLM grades each passage and keeps only the ones that answer the question, so the agent says "not found" instead of guessing.
- A human decides. Submitting an application or running a paid search pauses until the user approves that exact action. This is enforced in code, not in the prompt.
- Privacy by design. Address, phone and email are never sent to the LLM, and each user's data is fully separated.
Tech stack
LangGraph and LangChain, ChromaDB with local multilingual embeddings, OpenRouter (GPT-4o mini), FastAPI, Next.js, PostgreSQL and Tectonic for LaTeX CVs.
Evaluation
After fixing a retrieval gap, Ragas faithfulness rose from 0.82 to 0.91 and answer relevancy from 0.66 to 0.92. The agent picks the right tool for 91% of 34 labelled prompts in English and German, and a typical chat turn takes under 3 seconds, with the real cost of every LLM call tracked.
