How AI Is Used and Contained
- AI routing and the linguistic air-lock
AI routing in the PointSav platform processes language model requests through a local sanitization step before any data reaches external models, ensuring that internal structured data never travels to third-party servers in identifiable form.
- Decode-time constraints
Decode-time constraints are structural rules applied to a language model's output at each token-emission step, making banned vocabulary or structurally invalid responses mathematically impossible to produce rather than catching them after the fact.
- Doorman protocol
The Doorman is the sole AI request boundary through which every inference call routes — enforcing sanitise-and-rehydrate discipline once, logging every call to an immutable audit ledger, and capturing the training signal that compounds the platform over time.
- Elastic Compute #1 nightly LoRA training pipeline
Nightly two-phase pipeline on Elastic Compute #1 that rebuilds the deployment DataGraph and trains LoRA adapter weights for the workspace language model.
- Knowledge flow: training loop and ontological DataGraph
Quality framework for the Totebox knowledge flow, asking whether LoRA adapters measurably improve the model and whether the DataGraph is an accurate ontology.
- Learning Datagraph — SLM trajectory loop and apprenticeship queue
Four-leg training loop turning operator interactions into training tuples — trajectory capture, apprenticeship queue, editorial DPO pairs, and correction distillation.
- PointSav-LLM
The planned vendor-tier specialist AI model for substrate-sovereign SMBs — Tier 3 of the Four-Tier SLM Substrate Ladder, built by continued pretraining of OLMo 3 32B on the platform's federated apprenticeship corpus.
- SLM Rust stack architecture
The full Rust dependency graph and binary architecture for service-slm, the Doorman service that mediates every inference call in the PointSav platform.
- Tiered Entity Extraction Architecture
The PointSav entity extraction pipeline runs three tiers in sequence on each document: Tier 0 provides fast extractive detection via GLiNER; Tier A provides a generative fallback via OLMo on CPU; Tier B provides GPU enrichment and records improvements as training signal.
- Zero-container inference
Planned Tier B GPU deployment pattern using native Linux binaries under systemd, with idle-shutdown timers halting GPU billing when inference queues are empty.