Friday: eight hours of ideas
The team spent almost the entire first night discussing directions, constraints, and possible agent products before committing to a build.
HACKATHON BUILD
TickyTalky
TickyTalky is a self-improving AI content strategist for TikTok creators. Built by Giga-Pradhi Team for the AITX Community x NVIDIA Claw Agent Hackathon, with Red Hat AI involved and the team connection starting through Antler, it backtests creator ideas against real posting history and turns the learning into a next-week posting gameplan. The event ran three days, but after a team reset, the actual product build happened in less than 24 hours.
Proof
Public product and demo screenshots.
Timeline
A three-day hackathon, a sub-24-hour build.
The useful context is not just that TickyTalky shipped at a weekend hackathon. It is how late the final idea locked, how much the team changed mid-event, and how compressed the actual build window became.
Saturday morning: team reset
Two teammates dropped out and no-showed, forcing the remaining Giga-Pradhi Team to regroup quickly.
Saturday 2 PM: TickyTalky became the plan
David rallied the team around his TikTok agentic AI idea: a content strategist that could backtest creator ideas against real posting history.
Saturday 3 PM to Sunday 11 AM: build window
Code and presentation were due by 11 AM Sunday. The event ran three days, but the real product build happened in less than 24 hours.
Product
An AI strategist that learns from the creator's real history.
The demo uses a Student, Dealer, and Grader loop. The Student generates ideas. The Dealer reveals historical posts one at a time with no look-ahead. The Grader evaluates blind outcomes. The point is practical: learn what works for a real creator without pretending the system knows the future.
The cached demo for @technically.david shows the full loop: profile stats, historical video verdicts, warm-vs-cold learning, and a seven-day recommendation plan.
Demo
What the public walkthrough shows.
Cached analysis for @technically.david
Part of the TickyTalky hackathon demo and recommended creator strategy output.
Profile stats and historical posting context
Part of the TickyTalky hackathon demo and recommended creator strategy output.
Green and red video verdicts
Part of the TickyTalky hackathon demo and recommended creator strategy output.
Warm-vs-cold learning curve
Part of the TickyTalky hackathon demo and recommended creator strategy output.
Seven-day gameplan of recommended videos
Part of the TickyTalky hackathon demo and recommended creator strategy output.
Training System
Recursive learning without retraining weights.
- Student generates candidate TikTok ideas.
- Dealer reveals historical posts one at a time with no look-ahead.
- Grader evaluates blind outcomes against the creator history.
- DeepSeek-V4-Flash served from a single NVIDIA DGX Spark through patched vLLM in a Red Hat AI-adjacent infrastructure context.
- NemoClaw/OpenShell recursive harness converts failures into persistent Playbook lessons without retraining model weights.
Team
Giga-Pradhi Team.
Giga-Pradhi Team was David Bramante, Timothy Vang, and Asit Mahato. AITX Community brought together builders around applied AI agents. NVIDIA provided the infrastructure focus around Claw Agents, DGX Spark, vLLM, and Nemotron. Red Hat AI was part of the event context around open, production-oriented AI infrastructure. Antler was the ecosystem where the team connection started.
David Bramante
Frontend, UX, Firebase/Supabase/Vercel deployment, and public demo.
Timothy Vang
DGX Spark and patched vLLM model-serving path for DeepSeek-V4-Flash in the NVIDIA and Red Hat AI infrastructure context.
Asit Mahato
NemoClaw/OpenShell recursive agent harness with Playbook learning.
Next Hackathon
Need a teammate who can ship the demo?
CodexDeus builds working AI products, demo apps, agent workflows, and deployed MVPs for hackathons, founders, and practical software teams.