video1.0<iframe src="https://www.loom.com/embed/34a236e6e2e24d20bbbc0d1367276a24" frameborder="0" width="1670" height="1252" webkitallowfullscreen mozallowfullscreen allowfullscreen></iframe>12521670Loomhttps://www.loom.com12521670https://cdn.loom.com/sessions/thumbnails/34a236e6e2e24d20bbbc0d1367276a24-6e51381912deb54d.gif299.44TickyTalky is a self-improving AI content strategist for TikTok creators. In this demo, we show how the system backtests a creator’s real posting history with a Student, Dealer, and Grader loop: the Student generates ideas, the Dealer reveals historical posts one at a time with no look-ahead, and the Grader evaluates blind outcomes. The demo shows cached analysis for @technically.david, including profile stats, green/red video verdicts, a warm-vs-cold learning curve, and a seven-day gameplan of recommended videos. We also explain the technical training system: DeepSeek-V4-Flash served from a single NVIDIA DGX Spark through patched vLLM, plus a NemoClaw/OpenShell recursive harness that converts failures into persistent Playbook lessons without retraining model weights. Built by Giga-Pradhi Team for the AITX Community x NVIDIA Claw Agent Hackathon. www.TickyTalky.com