CoverCraft
A cover letter generation tool designed around the prevention of AI hallucinations. It employs mechanisms such as judgment-computation separation, GitHub commit record anchoring, and adversarial review to ensure that all generated content is grounded strictly in the user's actual résumé, rather than fabricated or embellished information.
Timeline (last 90 days)
CoverCraft is a cover letter generator whose core design principle is that AI cannot claim anything that cannot be traced back to the actual text of the user's real resume
In CoverCraft, the LLM is only responsible for classifying the match between a resume and a job posting as STRONG_MATCH / PARTIAL / MISSING, while the final score is calculated by the application code using a fixed formula
The match score calculation formula is: score = (strong*1.0 + partial*0.6 + transferable*0.4) / total
CoverCraft pulls the actual commit history of the user's public repositories via the GitHub MCP tool, anchoring capability claims to specific commit hashes
CoverCraft uses an adversarial review module to downgrade exaggerated verbs, for example replacing "spearheaded" with "led implementation of" and "pioneered" with "developed"
CoverCraft's company research information is presented to the user in card form, where the user approves or rejects each item individually rather than having it automatically injected into the prompt
CoverCraft's tech stack is Next.js + Python MCP server + Gemini (with fallback and circuit-breaker mechanisms for 429 rate limiting) + Tavily search + Langfuse trace monitoring
CoverCraft's entire system runs on the free tiers of various service providers, with an operational cost of $0
CoverCraft, registered as an independent MCP server, has been listed in M8ven's MCP Trust Index and marked as a Verified Publisher
Methods for systematically constraining generative AI hallucinations include stripping out deterministic calculations, anchoring facts with verifiable external data sources, filtering exaggerations through adversarial secondary review, and requiring human approval
All Facts (10)
Methods for systematically constraining generative AI hallucinations include stripping out deterministic calculations, anchoring facts with verifiable external data sources, filtering exaggerations through adversarial secondary review, and requiring human approval
50%UnverifiedIn CoverCraft, the LLM is only responsible for classifying the match between a resume and a job posting as STRONG_MATCH / PARTIAL / MISSING, while the final score is calculated by the application code using a fixed formula
50%UnverifiedThe match score calculation formula is: score = (strong*1.0 + partial*0.6 + transferable*0.4) / total
50%UnverifiedCoverCraft pulls the actual commit history of the user's public repositories via the GitHub MCP tool, anchoring capability claims to specific commit hashes
50%UnverifiedCoverCraft uses an adversarial review module to downgrade exaggerated verbs, for example replacing "spearheaded" with "led implementation of" and "pioneered" with "developed"
50%UnverifiedCoverCraft's company research information is presented to the user in card form, where the user approves or rejects each item individually rather than having it automatically injected into the prompt
50%UnverifiedCoverCraft's tech stack is Next.js + Python MCP server + Gemini (with fallback and circuit-breaker mechanisms for 429 rate limiting) + Tavily search + Langfuse trace monitoring
50%UnverifiedCoverCraft's entire system runs on the free tiers of various service providers, with an operational cost of $0
50%UnverifiedCoverCraft is a cover letter generator whose core design principle is that AI cannot claim anything that cannot be traced back to the actual text of the user's real resume
50%UnverifiedCoverCraft, registered as an independent MCP server, has been listed in M8ven's MCP Trust Index and marked as a Verified Publisher
50%