Launching today
Ghost in the Log
A Halloween log investigation with a live AI behind it
1 follower
A Halloween log investigation with a live AI behind it
1 follower
Something is haunting the logs. Pick one of three fictional incident cases and investigate together with a live AI: it walks the numbered log lines with you and lays out a possible cause, the evidence for it, and what stays unknown. The model never receives raw user input — only a preset scenario ID and a language. Logs are treated as data, never as commands, and every answer is validated (JSON format, evidence line numbers) before it is shown. Happy Halloween! 🎃
Free
Launch Team
AWS Activate CreditsCredits that fuel your startup’s next milestone
Promoted
Hi Product Hunt 👋 I'm Kentaro, a cloud architect in Japan. I built Ghost in the Log for a Halloween hackathon because I spend my days reading logs — and I wanted to see if a live LLM could help investigate them without becoming a risk itself.
So the AI here is deliberately boxed in: preset scenarios only, logs as data (prompt injection has nowhere to land), and a deterministic checker that validates every answer before it reaches the screen. The ghost can talk, but it can't touch anything.
I'd love feedback on two things: does the investigation feel fun, and is the guardrail design convincing? The full write-up (including the safety design) is on DEV, the code is MIT on GitHub, and there's a 90-second narrated demo on YouTube. Thanks for stopping by — and happy Halloween! 🎃