# Minutes of Meeting - Enterprise Brain Scrum **Date:** March 23, 2026 **Duration:** ~30 minutes **Source transcript:** [20260324.txt](20260324.txt) **Gmail source:** [Notes: "Enterprise Brain Scrum" Mar 23, 2026](https://mail.google.com/mail/u/0/#all/19d1dfa9890404f5) · Message ID: `19d1dfa9890404f5` **Google Doc:** [Enterprise Brain Scrum - Notes by Gemini](https://docs.google.com/document/d/1yMizFRTOw2tL-kOk5QaYczpQrnTeFnqYok1oYmbW3F8/edit?usp=meet_tnfm_email) **Attendees:** Vara Kumar Jagarapu, Rajashekar G, Abhilash Adunuri, Yeshwanth Reddy Yerraguntla, Kalakonda Harshith Rao, Manisha Gundapuneedi --- ## Act I — Data must stay rich until the very end (00:00 – 00:06) The meeting opened on a schema design question Abhilash had raised in a document. He articulated the core concern: if you summarize too early when aggregating data across sources, you lose the information needed to correctly identify which user can access what. The team converged quickly. > *"Summary is, in other words, the very last thing that you have to generate before sending to the user. Until then you try to maintain as much structure and richness in the data as you can."* > — Yeshwanth Reddy Yerraguntla _(~00:04:18)_ The rule is now clear: combine data from all sources and pass it straight to the RT snapshot. No intermediate summarization. When the final email is being composed for a specific user, only then does the agent summarize — in plain English grammar, with all bullet points, hashtags, user info, and privacy context preserved until that moment. The follow-on action was equally concrete: Abhilash needs to write the DB schema for both organisation and snapshot today, and Rajashekar will enhance the documentation in parallel. ## Act II — UI rendering model gets clarified (00:06 – 00:12) Vara Kumar escalated a question from Anita (working on React integration): how should the agent response be wired through to the front end? The team needed to settle the rendering contract. > *"We need to send the JSON payload to React. React itself figures out what needs to render — there will be UI visualization functions to determine what needs to be rendered."* > — Rajashekar G _(~00:10:01)_ Yeshwanth directed Anita to inspect the existing `make UI` repo, which already runs pylog with all three agents. The network tab shows how the front end maintains a WebSocket connection and pushes tokens. The short conclusion: the server generates structures; React consumes a JSON payload and runs local visualization functions to decide what to render — not server-side HTML. ## Act III — Deployment, branching, and Copilot Kit (00:12 – 00:15) Rajashekar ran through the sprint board. The experiential UI task and agent-setup-for-UI are both in progress. A DB308 ticket — "valid Copilot Kit for dashboard and chart integration" — surfaced as a gap: Anita has been working on it but her subtasks are not listed under the story, so they need to be formally added before the story can advance. > *"Enterprise brand plus experience center — we should enable or disable based on feature flag. That is how they wanted it."* > — Vara Kumar Jagarapu _(~00:15:35)_ On deployment, Rajashekar completed the three-into-one branch consolidation and is now implementing feature flags. The feature-flag scope covers both Enterprise Brand and Experience Center. Pawani's Gmail rollout story will be linked to this deployment branch, and once Rajashekar syncs with Pawani the combined work will go to Prabhakar for an initial smoke test before escalating to Naru. ## Act IV — Testing is blocked; the user-aware agent comes first (00:17 – 00:24) Yeshwanth flagged a concrete blocker: there is no user-aware agent yet, so the RT-agent-to-user-agent communication pathway cannot be tested at all. The second blocker — how to simulate multiple users — turned out to be premature. > *"Basically, we didn't create the user-aware agent yet to test that out. I'm also blocked on core agent knowledge base, arc snapshot — and the second blocker is needing multiple users. Or maybe I don't need multiple users to build that user-aware agent."* > — Yeshwanth Reddy Yerraguntla _(~00:19:02)_ Vara Kumar clarified the path forward: create a basic skeleton of the user-aware agent using mock data — a snapshot table with user and preference fields — filter by user in the query, and close the current ticket. End-to-end testing with real data and multiple users is deferred to the next sprint. For the current ticket, Rajashekar will scope it to mock-data test cases only. ## Act V — Alert lifecycle nears schema stage (00:21 – 00:28) Vara Kumar walked through Yolia's alert lifecycle implementation. The pipeline: sentiment scores are added to emails and Jira data at the data-source level → the RT snapshot stores this enriched data → the RT agent receives it → Yolia's code routes alerts to the user-aware agent filtered by user role → the UI displays alerts accordingly. > *"At data source level, whatever emails or Jira information we are receiving, she is adding sentiment scores. The RT agent receives data from the RT snapshot, and based on that she is going to send those alerts to the user-aware agent, which then sends to the UI based on roles."* > — Vara Kumar Jagarapu _(~00:22:32)_ Rajashekar clarified that Yolia is not on implementation yet — Abhilash and Amula are still finalising schema and table design. Hashith Rao, Rahul, and Rakesh own the alert lifecycle backend, org-snapshot knowledge graph, and data-source agent tasks respectively. On automation testing: Vara Kumar spoke with Prabhakar about GenHRX's approach. The recommendation is not to invest in heavy test automation now because iteration churn makes it a burden. The preferred path is to maintain a test-case list and use an agent to regenerate tests once a stable version ships. Vara Kumar will investigate an internal resource who did this for a sample website and report back next week. --- ## Decisions | # | Decision | Owner | |---|----------|-------| | 1 | No summarization until the final user-email stage; pass raw structured data through to the RT snapshot | Abhilash (schema), Yeshwanth (design) | | 2 | UI rendering contract: server produces JSON payload; React calls UI visualization functions | Rajashekar / Anita | | 3 | Enterprise Brand and Experience Center will be gated behind feature flags | Rajashekar | | 4 | User-aware agent ticket will be closed with a skeleton + mock data; real multi-user testing deferred to next sprint | Yeshwanth | | 5 | Heavy automation testing deferred; maintain a test-case list and generate via agent post-stable-release | Vara Kumar | --- ## Todos Write DB schema for both organisation and snapshot; enhance architecture documentation today.
Abhilash Adunuri Today (Mar 23) Create skeleton of user-aware agent using mock data (snapshot + user preferences table); close current ticket.
Yeshwanth Reddy Yerraguntla Current sprint Write test cases scoped to mock data only (real E2E deferred).
Yeshwanth Reddy Yerraguntla Current sprint Sync with Pawani on Gmail rollout implementation; review full story end-to-end; give one version to Prabhakar for smoke test, then forward to Naru.
Rajashekar G This week Complete feature-flag implementation (Enterprise Brand + Experience Center); link Pawani's Gmail story to the deployment branch.
Rajashekar G Current sprint Add Anita's subtasks to story DB308 (Copilot Kit for dashboard and chart integration) so they appear on the board.
Rajashekar G / Vara Kumar Jagarapu Today (Mar 23) Investigate agent-based test-case generation from the internal resource who worked on the sample website; bring a KT/update back next week.
Vara Kumar Jagarapu Next week