I built CoverCraft because I kept seeing a problem with AI-assisted job applications: fluent output can still contain claims that aren't actually supported by a candidate's experience. So I treated the cover letter as an evidence problem rather than a text-generation problem. CoverCraft currently separates: • Resume evidence extraction • Job-description analysis • Live company research • MCP-based tool orchestration • Human approval before final synthesis • Adversarial checks for overclaims and unsupported metrics • Langfuse tracing for execution visibility The core idea is simple: LLMs can classify and synthesize evidence, but deterministic application logic should control what is actually allowed into the final output. I’d especially love feedback from engineers working on RAG, agents, MCP, or LLM evaluation: where would you challenge this architecture? GitHub: https://github.com/Ambuj123-lab/...

CoverCraft AI

Evidence-grounded AI for accurate, traceable job application

Evidence-grounded AI for accurate, traceable job application