A fully interactive build cost model for Dynery v1.1 "1 Louder" — complexity scored 1–5, effort scored 1–5, hours estimated, cost calculated at $60/hr. Filter, sort, and explore across all 12 product epics and 7 implementation phases.
Janea Systems (Seattle) built and deployed Dynery's AI foundation for Dynery. The production stack is more sophisticated than most pre-seed startups — and it directly reduces the complexity of nine capabilities in the roadmap. The BART model is already in production. The H3 spatial index is already powering discovery. DynamoDB is already storing the Dining Passport. The vignette engine is not being built from scratch — it is being extended from a running system.
Nine capability scores revised downward because the existing stack eliminates the hardest parts of each build. All scores in the interactive model below reflect these revisions.
| ID | Capability | Original C | Revised C | Stack component that reduces it | Reason |
|---|---|---|---|---|---|
| F20 | Vignette AI pipeline — core | 5 | 4 | BART (distilled) | Model selection, training infrastructure, and inference pipeline already exist. Extension from classification to seq2seq narrative generation — not a ground-up build. |
| F42 | Annual Vintage generation engine | 5 | 4 | BART (distilled) | Still the most complex NLG task, but model risk is eliminated. BART foundation handles summarization natively — the Vintage is a long-form summarization + synthesis task. |
| F21 | Story Prompt AI — 24hr post-meal | 4 | 3 | BART zero-shot | Zero-shot classification already identifies meal context. Story prompts are conditionally templated from existing model outputs — classification → specific prompt generation is incremental. |
| F28 | Occasion-aware narrative | 4 | 3 | BART zero-shot | Occasion classification is already within BART's zero-shot capability. Conditional narrative framing by occasion type is a prompt-engineering layer, not a new model. |
| F25 | Dynery Ops floor dashboard | 4 | 3 | DynamoDB Streams | Real-time change events are built into DynamoDB natively. WebSocket API Gateway + Lambda → DynamoDB Streams is well-documented AWS architecture. No custom event bus needed. |
| F36 | Voyage Feed — social discovery | 4 | 3 | H3 Spatial Index | Social discovery feed builds on the existing H3 spatial queries. Peer vignettes within a hexagonal cell radius are a straightforward extension of the current discovery mechanism. |
| F55 | Dynery Away — travel city guides | 4 | 3 | H3 Spatial Index | H3 was designed for exactly this. City-level hexagonal clustering for travel dining guides is H3's native use case — no additional geospatial infrastructure required. Content is the work, not the infrastructure. |
| F48 | Personal Portfolio — dynery.com/@user | 4 | 3 | FastAPI + DynamoDB | The data already lives in DynamoDB and is already served via FastAPI. The portfolio is a Next.js SSR frontend rendering existing API responses — a frontend build over an existing backend. |
| F59 | Constellation curation algorithm | 4 | 3 | BART zero-shot | Restaurant scoring for Constellation tiers uses the same zero-shot classification pipeline already powering discovery. The algorithm is a scoring layer on existing model outputs, not a new model. |
The Janea Systems advantage: They are a Seattle-based high-performance computing firm (makers of Memurai, a Redis-compatible Windows implementation). They built Dynery's AI foundation and know the codebase. For the vignette engine extension — the highest-complexity item in the roadmap — the fastest path is extending the Janea Systems partnership rather than onboarding a new AI engineering team from scratch. The BART model, Lambda infrastructure, DynamoDB schema, and FastAPI patterns they built are the foundation everything else builds on.
Every capability is independently scored on two dimensions. Complexity captures how technically difficult the feature is to architect and build correctly. Effort captures how much work is required to fully implement, test, and ship it. The product of the two scores (C × E, ranging from 1–25) determines the hours estimate.
Hours formula: The C × E product score (1–25) maps to a low/high hours range. Score 1 = 8–16 hours. Score 25 = 480–960 hours. Cost = hours × $60/hr. Low estimate assumes experienced engineers, clear requirements, and no significant rework. High estimate includes integration complexity, QA cycles, and realistic scope creep.
Filter by epic, priority, or phase. Search by keyword. Sort any column. The totals update live as you filter.
| ID | Epic | Capability & description | C | E | Score | Priority | Phase | Hours range | Cost range |
|---|
Phases overlap intentionally — Phase 3 experience work begins while Phase 2 infrastructure is still in progress. The critical path runs through Phase 3 (Vignette AI + Experience Coach) which determines the v1.0 ship date.
Each phase is independently shippable and delivers measurable value. Phases are not waterfall — earlier phases provide the foundation but later phases can begin as soon as their dependencies are ready.