Activity

1. owlstonbusiness-svg

   added

   enhancementNew feature or requestNew feature or request

   on Oct 28, 2026
2. owlstonbusiness-svg

   added sub-issues on Oct 28, 2026

   • [Ops · WA · #511] Business WhatsApp enquiry agent #512
   • [Ops · WA · #511] Master Sheet tutor search + recommend #513
   • [Ops · CRM · #511] Client relationship memory #514
3. owlstonbusiness-svg commented on Oct 28, 2026

   

   owlstonbusiness-svg

   on Oct 28, 2026

   CollaboratorAuthor

   More actions

   Trigger screenshot — Denise So → Hauzton (2026-10-28 ~23:17 HKT)

   Denise (with Ryan): use AI agents to (1) handle customer enquiries and (2) search Excel sheets and recommend tutors — try on WhatsApp Business Account.

   Hauzton: already planning; matching may be possible but errors may happen.

   CEO follow-up: open linked WA Business improvement issues + client relationship memory (recognised/remembered; AI scans chat+profile per client).

   Children: #512 enquiry · #513 sheet recommend · #514 client memory CRM.

   
4. owlstonbusiness-svg commented on Oct 28, 2026

   

   owlstonbusiness-svg

   on Oct 28, 2026

   CollaboratorAuthor

   More actions

   CORRECTED — Gemini WA Concierge / Silent CRM / Matching (vs #511–#514) — 2026-10-29

   Source: 3 Gemini packs (hospitality schema · FOH/BOH epic · search_tutors hard-filter).
   Locks: Denise/Ryan WA ask + CEO relationship-memory · #511 epic · #512 enquiry · #513 sheet recommend · #514 client memory · WA SOP · fee #508 · Blue Tick #498 · Cap.57 independent tutors · draft-first / CEO tutor-ping rules.

   Verdict snapshot

   Theme	Call
   FOH vs BOH split + Memory Inject	ADOPT
   Silent extract from chat (no survey wall)	ADOPT
   Waiting-window micro-hook (1 soft Q after “we’re searching”)	ADOPT
   Hard filter (code) → soft rank (AI) ≤3	ADOPT (maps to #513)
   Human approve match drafts before parent send	ADOPT (Phase 1)
   FAQ auto-reply; matching draft-first	ADOPT
   Unified contact memory JSON shape (role + children + soft notes)	ADOPT shape · OPTIMIZE storage → Master Sheet Notes first, then DB/#459
   Issue categories as desk flags	OPTIMIZE — not auto-spam GitHub
   Voice-note Whisper in V1	OPTIMIZE → V1.1
   “Omotenashi / VIP dossier” tone in parent-facing copy	REJECT brand voice — keep MatchMax warm energetic EN/繁中; remember without creepy
   Gold / “Verified Badge” naming	REJECT — use Blue Tick / Tier 2 Verified per #498
   Invented T-402 IDs	REJECT — use live MM-T### / MM-C###
   Auto tutor pings / auto-send shortlists	REJECT without CEO/CS rules
   Unpark on-platform AI chat (#476/#440)	REJECT
   Internal “employee” WA schema as product	REJECT from public bot — staff ops only, Cap.57 careful
   Centre salary / employment placement	REJECT (parked #508 EA)
   Grade-guarantee language in memory or pitches	REJECT

   ---

   ADOPT (ship into #511–#514)

   1. Two-agent split

      • FOH: warm replies + FAQ + present shortlist; prompt gets Memory Inject.
      • BOH: async after idle; extract facts → upsert memory; propose friction flags (human confirms).
        Maps: [Ops · WA · #511] Business WhatsApp enquiry agent #512 FOH FAQ · [Ops · CRM · #511] Client relationship memory #514 BOH memory · [Ops · WA · #511] Master Sheet tutor search + recommend #513 search.

   2. Zero-friction collection
      Passive NLP from natural chat (80%) + one caring Q in the waiting window (15%) + post-lesson / friction triggers (5%). Do not open with a survey checklist.

   3. Universal traits (everyone)
      Preferred language/register (EN / 繁中 / mix) · cadence (hours, voice vs text) · relationship temperature (plain words, not creepy scores in parent UI) · past friction notes.

   4. Role schemas (trim to MatchMax)

      • Parents: child name(s), school, curriculum, subjects, district/mode, budget; soft: persona, child learning vibe, schedule quirks, siblings; history: past tutors + why ended.
      • Tutors: subjects/curricula/rates/zones/Active; soft: reply speed, teaching vibe, blackout windows; fee settlement reliability.
      • B2B centres: POC, volume, preferred delivery (WA cards vs sheet); centre spin ladder [Fee · #464 · Urgent] 6-Month Ownership Boundary + Retained-Only + Drop floor #508 (1.5/1/0.5) in ops notes — not salary track.

   5. Matching architecture ([Ops · WA · #511] Master Sheet tutor search + recommend #513)
      LLM → structured search_tutors params → deterministic filter on Master Sheet → AI ranks ≤3 with “why they fit” from memory → admin draft card for Denise/Hauzton/CS before parent send.
      Hard filters: Active · subject+level · curriculum · district/online · rate ≤ budget (+ optional small buffer) · exclude prior rejects.
      Soft rank: Blue Tick boost · alma mater · reliability/punctuality if we have data · vibe keywords · budget sweet spot.
      Zero-result fallback: relax budget / offer Zoom — ask parent, don’t invent tutors.

   6. Privacy / Cap.57 in pitches
      No phone / full legal name on first shortlist · Independent Tutor framing · phones after fee paid (existing SOP).

   7. Policy lines FOH may state
      Parents $0 platform fee · tutor admin = 1.5 (L1 100% + L2 50%) · short ≤5 → 20% package · tutors = independent · Blue Tick = staff-checked docs · no grade promises.

   ---

   REJECT

   Gemini idea	Why
   Rename product “Omotenashi engine” / VIP interrogation vibe	MatchMax voice already locked; remember warmly, don’t sound like a dossier
   “Gold Verified” / generic Verified Badge as copy	CEO #498 → Blue Tick (en + zh-HK)
   Fake IDs T-402, USR-8821 in specs	Live codes MM-T / MM-C / MM-C#### cases
   Auto-send match cards or auto-ping tutors	Standing: draft-first / CEO approval for waiting-case pings; ≤3 waves
   Auto-open GitHub issues from every WA vent	Noise; use desk Issue flag → human files GH if real bug (#262 privacy)
   Unpark #476 / #440 / #450 / site AI search #22/#121	Epic is WA + Sheet + CRM only
   Treat tutors as employees in schemas/prompts	Cap.57 · independent · EA salary track parked
   Store “we guarantee Level 7” style outcomes	Fee/safety SOP
   Full Slack-first admin path as required	Prefer existing Business WA team groups + Master Sheet; Slack optional later
   Phase-1 Whisper/voice as blocker	Nice; defer until text path works
   Invent reliability_score / teaching_style_tags columns as launch blockers	Map to existing Active/capacity/Notes first; add columns when Denise cleans sheet

   ---

   OPTIMIZE (MatchMax-shaped)

   1. Storage V1: write memory into Clients / Tutors Notes (or Description) on Master Sheet + optional contact_memory JSON later (Supabase / [Urgent · V1] Unified account + roles + child sub-profiles + hours ledger #459 child profiles). Don’t block on new table week 1.
   2. Remember brief for desk: 3–6 line card before reply (“Anka — SSC · IB Eng A EE · ~$500 · warm 繁中…”) — staff-only; never dump raw JSON to parent.
   3. Issue codes → desk tags (MATCH-MISMATCH, TUTOR-CONDUCT, BILLING-1.5, BOT-FALLBACK, PLATFORM-UX) linked to MM-C / MM-T / MM-C####; GH only when eng bug.
   4. Fee memory: private always 1.5 each; centre intros use 1.5 / 1 / 0.5 / $0 ([Fee · #464 · Urgent] 6-Month Ownership Boundary + Retained-Only + Drop floor #508) — teach FOH/BOH; hide salary track.
   5. Preferred display name ([Feature · V1] Preferred name = optional public display name #502) in memory + pitches when set; else tutor code.
   6. Confidence: show source cells (subject, rate, Active) on admin draft; Hauzton’s “errors may happen” → never silent auto-match.
   7. Roadmap trim:
      • P0: Sheet Notes memory upsert + remember brief ([Ops · CRM · #511] Client relationship memory #514) · FAQ FOH ([Ops · WA · #511] Business WhatsApp enquiry agent #512) · hard-filter shortlist drafts ([Ops · WA · #511] Master Sheet tutor search + recommend #513)
      • P1: waiting-window Q · reject-list exclusion · Blue Tick / alma soft rank
      • P2: Whisper · auto 1.5 check-in (after [V1 · #467] Scheduling + revenue/earnings engine (4 buckets) #461/[Feature · V1] Case ratings carousel + bilateral End Case #503 gates) · JSONB column
   8. Assignees: Ryan (AI/backend) · Denise (ops/bot flow + sheet columns) · Tim/Austin (invoice/fee desk) · Hauzton (match approve) — keep; don’t invent new roles.
   9. Scoring weights: treat Gemini +30/+25/+20 as starting proposal — Denise tunes after 20 real chats; don’t hard-code into product ToS.
   10. Gender filter: only when parent explicitly asks (ADOPT Gemini rule).

   ---

   Map onto existing issues

   Child	Take from Gemini
   #512 Enquiry	FOH prompt rules · FAQ auto · BOT-FALLBACK handoff · language mirror
   #513 Sheet recommend	search_tutors schema · hard filter · ≤3 · admin draft · fallback copy · privacy pitch rules
   #514 Client memory	Universal + parent/tutor/B2B soft fields · Silent Extractor prompt · Memory Inject · waiting-window · staff-only brief
   #511 Epic	Architecture diagram FOH/BOH · roadmap P0–P2 above · non-goals unchanged

   Eng paste (short)

   ADOPT FOH/BOH + silent CRM + hard-filter→≤3 draft. REJECT Gold naming, auto-ping, GH spam, unpark in-app AI. OPTIMIZE Master Sheet Notes first; Blue Tick; MM-codes; centre 1.5/1/0.5; desk tags not auto-issues.
5. owlstonbusiness-svg commented on Oct 28, 2026

   

   owlstonbusiness-svg

   on Oct 28, 2026

   CollaboratorAuthor

   More actions

   Children #512–#514 ACs now include the full OPTIMIZE list from CORRECTED. Epic gained FOH/BOH architecture note, assignees (Ryan / Denise / Tim·Austin / Hauzton), and P0–P2 roadmap checkboxes. REJECT list unchanged — not reopened.
6. owlstonbusiness-svg commented on Oct 28, 2026

   

   owlstonbusiness-svg

   on Oct 28, 2026 – with Grok (by xAI)

   CollaboratorAuthor

   More actions

   Refinement — CEO 2026-10-29 (fee framing across FOH)

   Applies to all children; no policy invent beyond standing locks.

   1. FOH FAQ auto-replies ([Ops · WA · #511] Business WhatsApp enquiry agent #512) must use the same fee framing as [Fee · #464 · Urgent] 6-Month Ownership Boundary + Retained-Only + Drop floor #508 / [Content · V1] FAQ accuracy pass + Tutor Playbook vs Guide/Help IA #515:
      • Parent pays the tutor’s rate.
      • MatchMax 1.5 = tutor’s slice of those intros (not an extra parent charge).
      • Centre ladder 1.5 / 1 / 0.5 / $0; salary track hidden.
   2. First tutor ping template lists fee before YES ([Ops · WA · #511] Business WhatsApp enquiry agent #512 paste-ready).
   3. Fee confusion (Tatum-class) → desk tag BILLING-1.5 ([Ops · CRM · #511] Client relationship memory #514); human confirms; do not auto-open GitHub; do not rewrite the ladder.
   4. Matching drafts ([Ops · WA · #511] Master Sheet tutor search + recommend #513) that mention fee use the same lines; Cap.57 / no grade promises unchanged.

   REJECT list on CORRECTED comment stays closed. On Hold issues stay On Hold.
7. owlstonbusiness-svg commented on Oct 28, 2026

   

   owlstonbusiness-svg

   on Oct 28, 2026

   CollaboratorAuthor

   More actions

   Consolidated for eng — sub-issues listed: #512 #513 #514. Parent epic map: see #518.
8. owlstonbusiness-svg

   changed the title [-][Epic · Ops/WA] Business WhatsApp AI desk — enquiries · sheet matching · client memory CRM[/-] [+][Epic · Ops/WA] Business WhatsApp AI desk[/+] on Oct 29, 2026
9. owlstonbusiness-svg

   added a sub-issue on Oct 29, 2026

   • [Trust · #511 · Urgent] WA+site AI flags, pre-match gates, drop matrix #529
10. owlstonbusiness-svg commented on Oct 29, 2026

    

    owlstonbusiness-svg

    on Oct 29, 2026

    CollaboratorAuthor

    More actions

    Child: Trust + WA AI flags + pre-match gates → #529
11. owlstonbusiness-svg commented on Oct 29, 2026

    

    owlstonbusiness-svg

    on Oct 29, 2026

    CollaboratorAuthor

    More actions

    AC add (Grok ADOPT): Lightweight case risk flags queue UI so the WA AI classifier (#529) has a place to land advisory flags even before full automation. Schema slice: #535.
12. owlstonbusiness-svg

    added

    urgentGood for newcomersGood for newcomers

    on Oct 29, 2026
13. owlstonbusiness-svg

    added a parent issue on Oct 29, 2026

    • [Epic · V1] Matching Ops Desk #517
14. keyframesfound

    added

    manualHandled personally by the owner — agents must not buildHandled personally by the owner — agents must not build

    on Oct 31, 2026