A full codebase deep audit turned into a release: 7 critical silent failures fixed (customer history, follow-up feedback linking, empty reports among them), reports now fully German and print-ready, installation around 6.3 GB smaller (16 → 9 models), ~30,000 lines of dead code removed — with zero breaking changes for v1 consumers.
Howzer v2.5.2-DE: full analysis envelope, faster analysis, long-text severity
The v1 analyze response now returns the full analyzer breadth — purely additive, no breaking changes. Independent analyzers run in parallel (17% lower median latency), the severity model was retrained for long texts, and churn alerts were re-tuned on real data.
Howzer v2.5.1-DE: in-house sentiment model, report delivery, more hardening
A maintenance release that strengthens three areas: an in-house German sentiment ML model, automated report delivery by email, and further robustness hardening for self-hosted operation.
A hardening release focused on reliability: honest results under load, customer replies grounded in the feedback, fail-closed persistence, air-gap-capable installation, and fully wired automation.
Enterprise intake: CSV import, channel normalization, and German text auto-extraction
The pipeline now auto-extracts PLZ, city, reference numbers, and urgency markers from German customer emails. Channels are normalized from 50+ variants. CSV batch import and post-hoc enrichment endpoints round out enterprise intake.
Production hardening release with 17 P0/P1 fixes across text preprocessing, database persistence, monitoring instrumentation, and Windows deployment compatibility.
The German-optimized pipeline is here, with improved sentiment analysis, emotion detection, root cause identification, and risk scoring. English pipeline remains on v2.2.