KnowledgeVault AI decomp
internal prototype · canonical JSON + Dreamborn Forge HTML
internal generated
decomp · supabase_json

KnowledgeVault AI decomp

decomp artifact · for KnowledgeVault AI · status approved

Planning Surface

Use this to decide what happens next.

Status

approved

open questions

No items captured.

Agent Handoff
Start Here

decomp artifact · for KnowledgeVault AI · status approved

Completion Evidence

No explicit evidence field yet. Require tests, screenshots, linked PRs, or reviewed outputs before marking complete.

Open Questions

No section body captured.

Structured Payload

Machine-readable source fields

phase
id

main

name

KnowledgeVault AI v1

status

approved

modules
idnamenotesfeaturessurfacesdepends on
M-00Data Model & MigrationsDeliver as numbered migration files. All other modules depend on this — it must be complete and reviewed before any other module starts.
M-01Auth & Expert Onboarding- F-01- S-01-a - S-01-b - S-01-c- M-00
M-02Product Catalog & Queuecoverage_pct = count(completed capture_sessions for this product) / products.target_session_count. capture_sessions table created in M-04 — stub coverage_pct as 0 in queue response until M-04 ships, or defer via view. product_pre_seeds table schema exists after this module; data populated by M-03.- F-02- S-02-a - S-02-b- M-01
M-03Pre-Seed Intelligence PipelineNo user-facing surfaces. M-04 cannot start until M-03 is shipped — POST /api/sessions/start requires a valid product_pre_seeds row for the selected product_id. Demo seed script is a required deliverable of this module. Supports F-02 (pre-session data) and F-03 (first_question at session start) as infrastructure — not itself a user-facing feature.- M-02
M-04Capture OrchestratorGPT-4o is the ONLY model in the real-time path — no o3 during session. WER compensation applied post-transcription. session_context JSONB written on every answered question (not just at close) for crash recovery. Does NOT include o3 evaluation, ground-truth gate, or payment creation — those are M-05. O3 and embedding costs appended to cost_breakdown in M-05.- F-03- S-03-a - S-03-b- M-02 - M-03
M-05Evaluation & Knowledge PublicationPayment record written ONLY after BOTH o3 scoring AND ground-truth gate return. knowledge_chunks published (is_published=true) only after pipeline completes — M-07/M-08 searchable corpus is empty until M-05 has processed at least one session. virtual_experts publish logic lives here — M-08 reads is_published flag but does not write it. O3 evaluation completes within 120s of session close.- F-04- S-04-a - S-04-b- M-04
M-06Payment & EarningsPayments created in M-05 (post-evaluation). This module owns release mechanism and payout UI only. Stripe Connect Express account created in M-01 — webhook handler added here. Can build in parallel with M-07 and M-10 after M-05 ships.- F-05- S-05-a- M-05
M-07Knowledge RetrievalCompany JWT middleware established here is reused by M-08 and M-09 — do not duplicate. field_queries table created here is consumed by M-09 (top_queries and slang_cloud in coverage dashboard). Ranking formula fixed for v1 — not configurable. Can build in parallel with M-06 and M-10 after M-05 ships.- F-06- S-06-a - S-06-b- M-05
M-08Virtual Expert QueryDepends on M-07 for company JWT middleware and text-embedding-3-small embedding setup. Depends on M-05 for virtual_experts publish threshold logic and is_published knowledge chunks. GPT-4o used here for answer synthesis only — not in evaluation path. Can build in parallel with M-09 after M-07 ships.- F-07- S-07-a - S-07-b- M-05 - M-07
M-09Coverage DashboardDepends on M-07 for field_queries table (top_queries and slang_cloud source data) and company JWT middleware. Depends on M-05 for knowledge_chunks (coverage_pct and slang_cloud). No real-time vector queries on load — sub-2s SLA achieved via materialized view only. Demo mode (company_id=demo) is a hard requirement for investor presentation. Can build in parallel with M-08 after M-07 ships.- F-08- S-08-a- M-05 - M-07
M-10Platform AnalyticsDepends on M-05 because cost_breakdown JSONB in capture_sessions is completed when o3 + embedding costs appended in M-05 (Deepgram + GPT-4o costs added in M-04). Admin JWT is separate from expert JWT (M-01) and company JWT (M-07) — simple Supabase Auth role=admin check. Can build in parallel with M-06 and M-07 after M-05 ships.- F-09- S-09-a- M-05
project

KnowledgeVault AI v1

strategy

11-module sequential-then-parallel split. M-01 through M-05 form a strict linear backbone — each phase is a genuine hard dependency for the next. After M-05 ships, M-06/M-07/M-10 unlock in parallel. After M-07 ships, M-08 and M-09 unlock in parallel. Key splits: (1) Capture Orchestrator (M-04) owns only real-time UI + Deepgram STT + GPT-4o interview conductor; Evaluation Pipeline (M-05) owns async o3 scoring + ground-truth gate + chunk publication + payment creation. (2) Pre-Seed Intelligence (M-03) isolated as pure-backend pipeline so Quinn can build Product Catalog UI (M-02) independently. (3) Knowledge Retrieval (M-07) establishes company JWT middleware and field_queries table that M-08 and M-09 both consume. M-00 (Data Model & Migrations) added as ground-zero — all modules depend on schema being stable before any feature work starts.

open questions

No items captured.

schema version

1.0

execution order
  • M-00
  • M-01
  • M-02
  • M-03
  • M-04
  • M-05
  • M-06
  • M-07
  • M-10
  • M-08
  • M-09
parallel groups
  • Item
    - M-06 - M-07 - M-10
  • Item
    - M-08 - M-09