AI in Mihwar
AI in Mihwar does the writing and reading — drafting, summarizing, converting — while the numbers stay deterministic. Every AI output is a draft a human reviews; nothing AI-written becomes a record until someone presses Save.
What it can do today
Write your operational documents
From Documents → Create with AI, Mihwar drafts six kinds of documents — SOPs, recipes, policies, training material, audit checklists, and HACCP plans — in English or Modern Standard Arabic, streamed section by section. You can regenerate any single section, edit everything, and save into the normal document library (versioning and read-and-sign included). Where a fact is genuinely unknown, the AI writes [TO BE FILLED BY MANAGER] instead of inventing one.
Draft checklist templates
From Checklists → Create with AI, describe the routine (or pick a starter kind — opening, closing, HACCP, receiving…) and the AI proposes a full structured template: sections, item types (temperature, photo, pass/fail…), critical flags, and a schedule. You preview it as cards, edit, then save — it lands as a normal template you refine in the standard editor.
Turn an SOP into a checklist
On any document with a written body, “Draft checklist from this SOP” sends it to the checklist workspace with the conversion already set up — the AI derives the steps, picks sensible item types, and marks food-safety-critical steps. The procedure your team signs and the checklist they run stay one click apart.
Investigate incidents with you
On a CAPA incident, “Draft with AI” writes the hard part: a 5-Whys chain, the systemic root cause, and the corrective + preventive actions — built from the incident itself, its activity history, and similar past incidents at that branch (including recurrence clusters). It fills the investigation form as an editable suggestion; the four-eyes close gate and the audit trail are untouched.
Brief the owner every morning
The dashboard opens with a short AI-written brief: what's on track, what slipped, and the one to three actions worth doing first — phrased from the same numbers the tiles show, in your language. It generates once per day (refresh it anytime) and simply doesn't appear when AI is off.
Forecast demand — without a chatbot
Ordering suggestions come from a separate, dedicated forecasting pipeline (statistical models trained on your order history), not from a language model — that's why every suggestion carries a confidence score and a checkable accuracy page. The language layer only explains; it never picks the quantity.
The rules every AI feature follows
- Drafts, not records. AI output lands in an editable form or preview. A human saves it — and the saved record carries the normal audit trail.
- No invented facts. Unknown specifics become “to be filled” markers; the morning brief may only use the numbers it is given.
- Numbers are deterministic. Counts, forecasts, and accuracy come from your data and the ML pipeline; AI phrases them.
- Everything is logged. Every generation — who, what, which model, tokens, and estimated cost — is on the usage ledger your admin can see.
- Arabic is first-class. Every generator produces English or Modern Standard Arabic; bilingual fields store the generated language in the right slot.
Admin controls & cost
Admins govern all of it from Settings → AI Assistant: turn AI on or off for the organization, pick the default model, allow or block individual capabilities (each template type can be disabled separately), and set per-user and per-org hourly rate limits. The same page shows a 30-day usage dashboard with generation counts and estimated spend. Generation typically costs fractions of a cent on the fast tier; the cost estimate is shown before long generations.
What's next
On the roadmap, in rough order: extracting expiry dates and details from photographed certificates (halal, food-handler cards) straight into the registers that gate POs and rosters; quiz-question generation from documents; structuring pasted customer complaints into CAPA intakes; photo sanity-checking on completed checks; and one-click translation of org-authored templates between Arabic and English. Each will follow the same rules above.