Ideas
This directory contains Alex-curated project ideas, research agendas, and system-design sketches that are durable enough to keep in the wiki but are not yet source ingests or topic syntheses.
For idea pages, Relation To Foundation TSFM Agenda tables describe intended
agenda closure: if the idea works, which slots it would close or partially close.
Current evidence status belongs in the Evidence and Missing pieces columns,
not in a downgraded verdict.
Collaboration
These ideas are open research and product directions. If one of them matches your interests, I would be happy to discuss it, collaborate on research, or work on use-cases together.
Ideas are not the bottleneck. Hands are. Time-series modeling should be moving at least as fast as vision, audio, and robotics.
- Email: [email protected]
- X: @chemeris
- Telegram: @alexanderchemeris
Files
- company-local-blockwise-fine-tuning.md - Research direction for adapting shared models on company-local data while exposing only bounded update signals.
- aionoscope-manifold-reconstruction-benchmark.md - Benchmark idea for reconstructing latent-variable manifolds inside time-series model representations instead of relying only on linear probes.
- dynamic-curriculum-jepa.md - Dynamic surprise-controlled curriculum learning for JEPA on large useful-signal-poor video and time-series corpora.
- hierarchical-modeling-fixed-flops-budget.md - Research design note for fixed-FLOPs adaptive hierarchy, budget regimes, and anti-collapse preservation across text, image/video, and time-series models.
- llm-agents-need-action-conditioned-world-models.md - Why LLM tool-calling agents are strong DevOps baselines but need action-conditioned world models.
- otel-control-minimal-demo.md - Minimal Kubernetes/OpenTelemetry demo slice for action-conditioned telemetry prediction, intervention ranking, and what-if UI.
- k8s-otel-control-gym.md - Kubernetes/OpenTelemetry control gym for action-conditioned DevOps world models.
- tsl-jepa.md - Query-conditioned JEPA interface for time-series retrieval, alerting, captioning, and structured readouts.