~/meetings / 2026-08-06
Meeting #16

> 2026-08-06

Guest session with Sean, an HEB software engineer working on the Flor platform (and creator of Fresco, HEB's fresh-production platform). Yohannes led an extended Q&A covering the SRS-to-Flor migration, HEB's internal data and auth architecture, and how Sean's team uses AI in code review and testing. After Sean left, Whit and Yohannes picked back up on Project Big Mart, locking in the game's MVP scope, its core task list and paranormal complications, and a first pass at the codebase's component architecture.

3 attendees(Whit, Yohannes, Sean (guest))
##topics covered
  1. 01.Sean introduced as a guest — an HEB software engineer working on Flor, previously led development of Fresco (HEB's fresh-production platform)
  2. 02.SRS-to-Flor migration explained — SRS and Flor are separate systems that don't sync; a manual order placed in SRS won't show as an active order in Flor
  3. 03.Flor's scope is expanding from pure inventory management into a platform for all of center store; shelf-edge/planogram tooling (Pog PSA) is rolling into Flor within about 4 weeks, pilot stores first
  4. 04.Core platform challenge: onboarding other departments' teams onto Flor without impeding their existing workflows, while fixing how authorization currently works
  5. 05.Current auth described as client-side only, with no server-side validation or ability to revoke access remotely — a fix (shared RBAC, role-based access control) is in progress
  6. 06.Internal lingo — SRS referred to as a 'mythical' legacy system; HEB still runs some systems on an IBM mainframe rented from Oracle
  7. 07.IAM ('AIM') team controls shared auth tooling like 1Password via Terraform, split into an applications repo and a teams repo so ownership can change without touching the client surface
  8. 08.Fresco walkthrough — Sean's largest personal project: user preferences (a first across HEB's internal platforms), an event/holiday production planner, and the Hub and Spoke system (designated 'hub' stores produce goods like bakery items and ship them to associated 'spoke' stores)
  9. 09.Fresco's production-tracking screens (replacing the old FPP screens) log who started a production run, at what store, and when, primarily for planning against forecasted demand — with daily production profiles that vary by time of day (e.g. ramp up before lunch rush)
  10. 10.Hyperdrive handbook — HEB's internal standards/guide library for engineers; Sean is a maintainer and wrote the localization guide and tooling (npx pull/push-translations commands, Slack notifications via the Localize third-party service)
  11. 11.Adoption strategy for new platform features — lower the barrier to entry, make it easy to try for free, then build mindshare with directors/APs once teams see the value
  12. 12.Inventory system complexity at HEB — four dedicated teams (Tim Classic, Tim Accounting, Evaluation, IO); migrating the legacy system to a new one called NIO built on event sourcing
  13. 13.CQRS (command query responsibility separation) explained as a pattern for scaling read-heavy vs. write-heavy clients by splitting into separate read/write databases
  14. 14.Event sourcing, write-ahead logs (WAL), and ACID transactions explained as the mechanism keeping a write database and its read replicas in sync without downtime
  15. 15.HEB's fragmented data ownership as a running problem — no company-wide data strategy; recipe data specifically is scattered across OVPS and D365 with no consumable API, which is why Fresco built its own recipe service to reconcile it
  16. 16.HEB actively moving away from direct database read/write access between teams and toward API-based access, though legacy systems make this a slow, coordination-heavy transition
  17. 17.HEB's AI tooling — the company evaluated Claude, Gemini, and GitHub Copilot and standardized on Copilot for its access to multiple models at lower cost; guidance has shifted toward planning with frontier models and implementing with smaller/cheaper ones as pricing has changed
  18. 18.AI code review — Sean's team built an AGENTS.md-style guide (called 'broth.md', for their internal AI reviewer 'Broth') to scope what the bot flags and avoid noisy, unhelpful reviews
  19. 19.Discussion of a cautionary story about a company where engineers stopped reviewing each other's PRs and let AI review AI-generated code end to end, and the broader worry about AI eroding human-to-human collaboration between roles (e.g. designers and engineers)
  20. 20.Sean's personal AI use — a personal knowledge-base agent ('Hermes') that tailors his resume per job application without altering the original, and a same-day rebuild of his team's test strategy (Storybook for unit-level, Playwright for end-to-end) that leans on AI once strong documentation and guardrails are in place
  21. 21.HEB partner feedback channels — a Slack 'Store P'/Flor feedback survey workflow, and an internal 'Flor to Feature' suggestion pipeline where product managers do read and respond to submissions
  22. 22.Sean invited to keep coming back to future SSI meetings
  23. 23.Project Big Mart MVP scope locked: one map, single-player first, five starter tasks — stocking, cleaning spills, checking out customers, hanging coupons, and moving pallets
  24. 24.Design goal for MVP is to nail the core gameplay loop — rising pressure from haunting events balanced against task completion — before adding upgrades or perks
  25. 25.Paranormal complications mapped to tasks: spills can be ectoplasm instead of the mundane version, haunted customers cause chaos and must be located (via clues) and exorcised, and power outages send the player to investigate the electrical room
  26. 26.Stamina reworked from a per-movement-depletion bar into a slowly-draining sanity-style meter that drains faster while sprinting or during certain paranormal events; all tuning values (drain rates, speed multipliers) to be config-driven
  27. 27.UI plan: a circular haunting meter, task list, shift clock, stamina bar, and a simple single-slot rotating hotbar (Resident Evil/Minecraft-inspired) for carried items like a flashlight
  28. 28.MVP win/lose condition simplified to a single shift: survive to win, lose if the haunting meter fills; multi-shift structure deferred until after MVP
  29. 29.Lunch break and perk systems intentionally cut from MVP scope (a suggestion Yohannes got from Claude) to keep the core loop tight: task, paranormal/ghost event, task
  30. 30.Component architecture discussed: Player as an orchestrator with Godot node-based components (using @export slots), GameManager for state/win-loss/haunting triggers/task lists, an Interactable pattern (proximity prompt → action), UI, and a shared tunable data-config file
  31. 31.Collaboration workflow settled on for Big Mart: single shared repo with feature branches and PR-based review, rather than per-person forks
  32. 32.Testing approach for MVP will be manual playtesting for now, with a debug/god-mode menu considered later for feature testing
  33. 33.Yohannes to build player movement and a first 3D prototype scene the same day
##decisions & next steps
  • →Sean invited to return as a recurring guest at future SSI meetings
  • →Project Big Mart MVP scope finalized: single map, single player, five starter tasks each paired with a paranormal complication, one-shift win/lose condition
  • →Lunch breaks and perks explicitly deferred out of Big Mart's MVP scope
  • →Big Mart collaboration model set: one shared repo, feature branches, PR review — no per-person forks
  • →Yohannes to start on player movement and a first 3D prototype scene right after the meeting
##notes
Transcript was unlabeled; attendees and per-speaker attribution were inferred and confirmed with Whit before writing.
← meeting #15