Block a user
research: LLM prompt strategies for tip generation quality
feat: LLM tip quality monitoring dashboard in admin
Todoist sync via events (not on-demand fetch)
Shipped in 2a73809 and verified in deployment 2026-04-18: NATS bridge connects, signals/feedback streams created, scheduler emits signals.task.synced on the 15-min tick. Smoke test passed;…
Todoist sync via events (not on-demand fetch)
events: NATS JetStream ingress + normalized event schema
events: NATS JetStream ingress + normalized event schema
Shipped in 2a73809 and verified in deployment 2026-04-18: NATS bridge connects, signals/feedback streams created, scheduler emits signals.task.synced on the 15-min tick. Smoke test passed;…
ml/serving JetStream consumer for signals.> + feedback.>
refactor: generalize recommendation pipeline — candidate → rank → render stages
feat: tip kind system — task, advice, insight, reminder
feat: TipCandidate shared schema — typed candidates through the bandit pipeline
feat: LLM output validation + structured JSON retry for tip generation
feat: prompt versioning — track prompt_version + model in tip_scores
feat: context assembler — user signals → structured LLM prompt context
infra:
ai compose profile — Ollama + LiteLLM for local dev
feat: AI gateway — wire ml/serving to LiteLLM with model aliases
feat: AI tip generation via Ollama — contextual advice from user signals
bug: reward updates to ml/serving silently swallowed on failure