I Can Generate Faster Than I Can Think: Tokenmaxxing, verification hell, and the creativity trap of infinite AI
- tokenmaxxing
- agentic coding
I used to think the bottleneck was execution.

Calvin Ku
@calvinku
Portfolio ©2026
Machine learning practitioner with 13+ years bridging research, product development, and technical communication. Background spans medical AI (Nature Communications paper on gastric cancer detection), financial ML (credit risk modeling that outperformed FICO; wine asset pricing using collaborative filtering), and applied NLP / agent systems.
Combines deep ML hands-on experience with public speaking at major conferences (WAIC 2018, ICCV 2019), open-source contribution (MemFuse, Chat2Graph), and product-building experience as co-founder of multiple ventures. Equally comfortable writing production ML code, explaining technical concepts to non-technical audiences, and building developer-facing content.
Currently active on Threads (2,000+ followers, growing) in AI / agent memory niche.
Talk on network effects & startup cold-start: youtube.com/watch?v=BgjXVKvscVg

Co-founder · Architecture & Python SDK
Open-source memory infrastructure for LLMs with cognitive architecture, intelligent buffering, and unified search—solving the statelessness problem that plagues every AI application

Creator & Developer
A focus timer that ties your Pomodoro sessions to a GitHub repo — shipped and live at tomopomo.com, with an ~80% session-completion rate across a modest, organic user base.
I used to think the bottleneck was execution.
In 2023, almost everyone thought agent memory was mostly a context-window problem.
My mentor survived three part-time jobs and bone-deep poverty with a simple tool: the Pomodoro technique. It worked for him. And for years, it has worked…