Fig. 03 — One ink: Teal
Projects
| Shape | ID | Title | Blurb | Year | PC1 | PC2 |
|---|---|---|---|---|---|---|
| P-01 | Agent Ledger: Shared Memory for AI Agents | A shared memory layer that lets web agents learn from past failures, and a benchmark showing when retrieval helps and when it hurts. | 2026 | 0.51 | −0.55 | |
| P-02 | GitXL: Version Control for Excel | Git for Excel workbooks: cell-level diffs, three-way merges, and a normalizer so re-saving an unchanged file produces no diff. | 2026 | −0.28 | −0.05 | |
| P-03 | Monte Carlo ICU Risk Model | An absorbing Markov chain fit to 14,000+ ICU transitions, with a Monte Carlo engine for tail risk in mortality and length of stay. | 2026 | 0.33 | 0.27 | |
| P-04 | Glitch Market | An on-chain marketplace for speedrun glitches, with encrypted delivery, decaying prices and jury-settled disputes. | 2026 | −0.29 | −0.49 | |
| P-05 | ECG Classification with a 1D CNN | A 1D CNN that flags abnormal ECGs at 0.98 AUC and uses Grad-CAM to show which parts of the heartbeat drove each call. | 2026 | 0.32 | 0.65 | |
| P-06 | Equities Backtesting Dashboard | A Streamlit dashboard for backtesting EMA-crossover and mean-reversion strategies, stress-tested with Monte Carlo resampling. | 2026 | 0.09 | −0.06 | |
| P-07 | Skate-Tracker | A skateboarding progression app: a prerequisite graph of 40+ tricks, RAG-based coaching, and video comparison against pros. | 2025 | −0.14 | −0.12 | |
| P-08 | setos.ai Startup Prototype & Pitch | A startup prototype that turns a plain-language question into a reading path through more than a million research papers. | 2025 | 0.05 | −0.56 | |
| P-09 | ForgeQB | An adaptive quizbowl study app with 500+ users, recommending from 160,000+ questions clustered into micro-categories. | 2025 | −0.54 | −0.10 | |
| P-10 | Inter-Agent Misalignment: Probing & Activation Steering | Testing whether a misaligned coding sub-agent can pull its conductor LLM toward reward hacking, and steering it back through its activations. | 2026 | 0.52 | −0.74 |
Things I built to see whether an idea holds up, from version control for Excel to shared memory for AI agents. Most pair a working product with a measurement of whether it actually works. Grey marks are the other five layers, kept in place so positions stay honest.