Demo — every team’s progress

Shipped / shown

• Automatic alerts + multiple Grafana dashboards.
• DLQs in many pipeline places, plus dead-letter replay.
• Symphony connector (chat app ≈ Slack) deployed to dev on a sandbox account — minutes before the call. Omar Ibrahim built it mostly alone.
• Symphony Grafana: webhook activity, throughput, message dedup, DLQ alerts. Empty of real data yet (just deployed).
• Event-router dashboard on staging + dev: some routed events, no errors / empty DLQ, freshness between router and connector.
• Microsoft connector dashboard: connector health + time since last event (idle because nobody is using it in that env).
• First-draft Microsoft connector onboarding runbook: client-side permissions + what goes in connector config. Share with product for “is this followable?”

Not done / next

• Symphony live demo slipped to next week (deployment kinks).
• Runbook is v1 — may need simplifying.
• Barry: in parallel to the long operating runbook, extract the access points to ask for early in client IT conversations (what to expose). He’ll work with Randa.
• Internet drop mid-share; Barry said ~95% of the talk survived; they missed the runbook screen until she reshared.

How the three layers fit

1. Ranking + context graph = knowledge layer. Streams from email / SharePoint / NetSuite → related tasks linked into one graph (entities, relations, evidence, decisions/rationale).
2. Background agent = automation. Takes linked events, checks task preconditions. If not satisfied → mark and pass to a deterministic state machine, which picks the next state and notifies the right owner or reviewer.
3. Copilot = chat. Ask about the process, email context, task DB (status, blockers, why blocked, what’s happening). Answers from that context, not generic chat.

Copilot demo (this week’s actual code)

• Guidelines rewritten with Barry + Banda after product feedback: less extra fields, less dwelling on completed-in-the-past work, more future-facing, more human wording, process transparency.
• Old “any at-risk tasks?” dumped everything including completed-but-late. New: only in review or in progress (not completed / cancelled). Status labels title-cased instead of raw DB IN REVIEW.
• Same filter on team-task questions: skip cancelled/completed.
• Staging by end of day — anyone can poke it and send feedback.
• Still building a privacy / ACL layer: who can see which tasks / emails.

Two surfaces — don’t mix them up

foundations.dev.clearsky.ai (as spoken). Reach out to Karino for creds. Some nav items will be hidden later — product + UX still picking what ships vs delays.

This week under the hood (not very demoable)

• PMO: notifications by role (reviewer gets ping when their tasks change); more credit to signal ranking.
• Karim + Noor: debugging, integration tests, TDD.
• Noor already started wiring wireframes → existing Cockpit APIs.
• Coming: Mixpanel, session replays, FDE events (FDE events are not Cockpit-core; extra instrumentation).

Web availability

• Azure Application Insights (non-prod): tests for Cockpit and other app URLs on dev + staging (ASR also said “JQL” and “HT5 / 72 app” — treat those labels as uncertain).
• Probes from multiple geos (East US, North US, …).
• 100% up in the SLA view they showed; history + failures available.
• One severity example on screen was wrong (showed Sev1 / outage that wasn’t real). Severities still being tuned.

Kafka = event backbone

• If Kafka is sick, customer data does not flow. Zidane credited for the alerts.
• Alerts: cluster not alive, redundancy, lag, consumer health.
• Grafana per env (dev shown: disk free on active controllers, lag, messages/min). Staging copy exists; plan is one dashboard per environment / client.
• Slack/mail alerts carry labels: env (dev/staging/prod), owner, extras.

The four weeks

Week 1 starts this Sunday (from 20 Aug → Sunday 23 Aug 2026). Each week already has intended deliverables. Next sprint tickets should already match those goals; some people already moved tickets.

Same content exists as Notion (Sherif says more readable) and a web/HTML export.

Risks / assumptions (called out on purpose)

• Internal vs customer vs vendor (named: CRAFT and the UI/UX firm).
• External deps tagged with “when we need them for this plan to hold.”
• v1 plan. Evolves as unanswered questions get answers. Mohamed asked: is this a live Linear dashboard? No — static report, Linear-fed via prompt, not an app. Could become a dynamic dashboard later; Mohamed wants that.

What they said

• Julia / Jiayi — Year 3, NUS. Accounting module was high-level. Reading SOX walkthroughs + journal-entry vocab with Ojas.
• Ojas — Year 4 computer engineering, NUS. First time US East Coast. Looking at workflows + where agents help; uses Hermes personally.

What leadership assigned (new to them on the call)

• Barry: (1) discovery tools — client artifacts → usable schema, semi-automated. (2) automated testing / QA tools — “they don’t even know this yet.”
• They’ll meet squads in coming weeks (today’s intros were for name→face).
• Mohamed: NUS Overseas Colleges — ~full-time year at a startup, soon-to-graduate. Batch of 9 interviewed → these 2. Stood out on agentic builds (Ojas’s tracking app: almost no hand-written code).