Working Title: Strata | Version: 2.1 | Date: April 2026
- Vision & Problem Statement
- Target Audience
- Core Features
- Content Pipeline
- Technical Architecture
- Feature Specs
- Subscription Model
- Monetization
- MVP Scope
- 8-Week MVP Launch Roadmap
- Risk Assessment
- Go-to-Market
- Platform Roadmap & Integrations
Hallow's 2026 Pray40 challenge featured The Brothers Karamazov and exposed 1.6M+ Catholics to the book at a surface level. That creates an immediate opening for Strata to become the deeper guided experience while awareness is still warm.
- Primary MVP book for launch
- Launch message: "The deeper dive after Pray40"
- Goal: Prove the pipeline and product loop on one culturally relevant flagship title before broad catalog expansion
- Public domain — available via Project Gutenberg
- Dense enough to showcase commentary, abridgment, notes, and quotations
- Strong overlap with Catholic, homeschool, Great Books, and sleep-listening audiences
- No high-quality guided podcast or YouTube version currently exists
- Ship Brothers Karamazov first as the flagship MVP
- Use America 250 as the next major content wave (July 4, 2026)
- Expand into Catholic Fathers, Stoics, and Scripture after pipeline is proven
Public domain literature contains some of the most transformative books ever written. Most people never read them because they are long, dense, archaic, and intimidating. Existing audiobook platforms use generic narration, no commentary, and no interactivity. Strata turns the world's greatest public domain books into guided, immersive audio experiences — commentary, abridgment options, contextual atmospheres, and AI-powered Q&A — at a fraction of the cost of licensed content platforms.
The gap: 1.6M people just did Hallow's Pray40 Lent challenge with The Brothers Karamazov as the anchor book. Hallow gave them surface-level reflections. Strata gives them the full guided deep dive.
- ADHD listeners who want structure and guidance through long books
- Catholic and Christian readers seeking formation and devotional content
- Homeschool families using Great Books curricula
- Intellectual self-improvers who want to finally finish the classics
- Philosophy and history enthusiasts
- Sleep listeners who use calm long-form audio
- College students and lifelong learners
- Anyone who saw a movie adaptation and wants to read the source
| Feature | Description |
|---|---|
| Guided Reading | Narrated episodes with embedded commentary every few minutes |
| Dynamic Abridgment | User selects depth: Full, Standard, Essentials, or Highlights Only |
| Atmospheres | AI-generated contextual ambient audio under narration |
| Multi-Voice Characters | Different AI voices per character in dialogue-heavy books |
| Sleep Mode | Slowed narration, extended pauses, sleep timer, resume-from-sleep |
| Focus Mode | Commentary off, pure narration at 1.25x–1.75x speed |
| Must Read / Most Quoted | Tags for landmark passages; filterable |
| Notes & Highlights | Timestamped notes synced to audio positions |
| AI Chat (Book-Specific) | RAG-based Q&A trained on the processed book + commentary |
| Voice Selection | Pro users choose from 10,000+ ElevenLabs voices |
| Book Requests | Pro users submit any public domain book for pipeline processing |
- Ingest — Pull public domain text from Project Gutenberg or Standard Ebooks via URL
- Segment — Claude splits text into 3–7 minute audio segments
- Analyze & Tag — Claude tags each segment:
type: narrative / philosophical / theological / biographical / dialoguescene_mood: battle / nature / sacred / contemplative / melancholy / triumphant / neutralatmosphere_prompt: natural language prompt for ElevenLabs Sound Effects APImust_read: booleanmost_quoted_passages: arraycharacters: named characters in segment
- Generate Commentary — the LLM writes 60–120s commentary per segment from the full episode text, then an automated verification pass checks it references only what's in the transcript; it enters an admin review queue and is served to users only once approved (see §6.4a)
- Abridgment Profiles — the LLM generates Full / Standard / Essentials / Highlights versions
- Narration Render — ElevenLabs voices the narration and commentary
- Multi-Voice — character voices assigned for dialogue segments
- Atmosphere Generation — ElevenLabs Sound Effects API generates an ambient clip per segment
- Assembly — FFmpeg mixes narration + ambient audio into the final episode
- Free tier: pre-mixed single file
- Pro tier: two separate files (narration + atmosphere) for client-side mixing
- Metadata & Storage — chapter markers, tags, timestamps, transcripts stored in Postgres; audio written to the object-storage driver (durable, private)
- Distribution — YouTube upload, self-hosted podcast RSS (free episodes), in-app streaming, and the public catalog API for Haven
Orchestration is Inngest, not n8n — each step above is a durable, retryable Inngest step with pause/resume/extend and onFailure cleanup.
This reflects the actual, implemented stack (as of the July 2026 audit and the Tier 0 work that followed). Earlier drafts of this PRD described an n8n / pydub / Supabase-Auth design that was never built; that has been removed.
| Layer | Stack |
|---|---|
| App | Next.js 14 (App Router), TypeScript, Tailwind CSS, Web Audio API |
| Auth | Custom JWT (jose) + bcrypt. One getProfileFromSession() path keyed on users.id. No third-party auth provider. |
| Database | PostgreSQL via Drizzle ORM (pg pool). shared/schema.ts is the source of truth; Supabase Postgres in production. |
| Audio storage | Storage-agnostic driver (src/lib/storage/): Supabase Storage or Cloudflare R2 in prod (private bucket + signed URLs), local disk in dev. Served only through a tier-checked stream endpoint. |
| Pipeline orchestration | Inngest (per-episode steps, pause/resume/extend/retry, timeouts, onFailure cleanup) |
| Audio assembly | FFmpeg (fluent-ffmpeg) — narration + atmosphere mix, duration probe, video shorts |
| AI — Segmentation & Commentary | Multi-provider LLM (src/lib/pipeline/llm.ts): Anthropic / Gemini (approved for flagship content), Groq / OpenRouter optional |
| AI — Narration, Voices, SFX | ElevenLabs or Microsoft Edge neural voices (free) behind a TTS driver; SFX/atmosphere always ElevenLabs |
| Commentary trust | Full-text grounding + automated verification pass + admin review gate + interpretive disclaimer |
| AI — Book Chat | Grounded Q&A over the processed book text (Pro) |
| Podcast distribution | Self-hosted RSS (Podcast 2.0), gated to free-preview episodes |
| Public catalog | /api/haven/catalog — CORS-friendly JSON for the Haven integration |
| Video distribution | YouTube (short-form) |
| Payments | Stripe (webhook-idempotent; entitlement read from the DB, not the JWT) |
| Deploy | Portable Next.js — Vercel / Netlify with no adapter; Cloudflare Pages via @cloudflare/next-on-pages. CI (typecheck/lint/test) gates every merge. |
- Chapter-by-chapter navigation
- Commentary toggle (on/off)
- 10-second skip forward/back
- Playback speed: 0.65x – 2x
- Bookmark any timestamp
- Chapter completion tracking
| Profile | Description |
|---|---|
| Full | Complete text, all commentary |
| Standard | Minor cuts to pacing, full commentary |
| Essentials | Core narrative + philosophical moments only |
| Highlights Only | Must Read passages + Most Quoted only |
- 10,000+ ElevenLabs voices
- Filter by gender, accent, age, tone
- Voice preview before committing
- Saved voice preference per book
- On-demand render triggered on selection
- RAG pipeline over processed text + commentary
- Grounded responses — no hallucination outside the book
- Example queries:
- "What does Alyosha believe vs Ivan?"
- "Explain the Grand Inquisitor scene"
- "What does Dostoevsky mean by active love?"
- Chat history saved per book
- Pro users submit any Gutenberg URL or title
- Pipeline processes automatically within 48–72 hours
- User notified on completion
- Approved books added to public catalog
- Claude tags landmark passages at processing time
- Users can filter episode list to Must Read only
- Most Quoted passages surfaced in UI with attribution
- Shareable quote cards (image + audio clip)
- Tap to highlight any sentence during playback
- Voice-to-text or typed annotation
- Timestamped to audio position
- Exportable as:
- Markdown
- PDF with timestamps
- Notion / Obsidian import
- Email digest
- Commentary disabled
- Speeds: 1.25x / 1.5x / 1.75x
- Minimal UI — only progress and skip controls
- Auto-advances through chapters
- Download full book or individual chapters
- Offline playback with full player functionality
- Sync notes when reconnected
- Cross-device sync via Supabase
- Resume from exact timestamp on any device
- Reading streaks and completion badges
Description: AI-generated ambient soundscapes that match the scene or mood of the content being read, mixed under the narration to create an immersive listening experience.
Pre-mixed single audio file — narration + AI-generated ambient audio baked together at ~15-20% ambient volume during pipeline assembly. Zero extra cost per stream.
Pro users receive two independent audio streams synchronized via the Web Audio API:
- Narration track — independent volume slider
- Atmosphere track — independent volume slider (0 = off, replace, or swap)
Atmosphere presets users can swap to:
- 🌧️ Rain / Storm
- 🌊 Ocean / Water
- 🔥 Fireplace
- 🌿 Forest / Nature
- 📻 Brown / White / Pink Noise
- ✝️ Gregorian Chant (loop)
- 🎵 Lo-fi / Study Music
- 🤫 Silence (narration only)
Claude tags each segment with:
scene_mood: [battle / nature / sacred / domestic / melancholy / triumphant / contemplative / neutral]
atmosphere_prompt: "medieval swords clashing, iron armor clanging, distant crowd noise"
ElevenLabs Sound Effects API generates the ambient clip per segment at processing time. pydub mixes it at episode assembly. Pro users receive the narration and ambient as two separate stored files, synced client-side.
Cost: Generated once at processing time per segment — no per-stream cost on either tier.
Differentiator vs. Hallow: Hallow's background music is licensed from a third party and is completely decoupled from content. Strata's atmospheres are contextually matched to the scene, AI-generated, and royalty-free.
Description: For books with dialogue and multiple named characters, ElevenLabs Studio auto-detects characters and assigns a unique voice per character, switching voices per dialogue line.
- Powered by ElevenLabs Studio API
- Available as a Pro feature
- Particularly effective for novels, plays, dialogue-heavy theology
- Pipeline passes character-tagged text to ElevenLabs Studio
| Setting | Normal Mode | Sleep Mode |
|---|---|---|
| Voice | User's choice | Slowest, warmest available voice |
| Speed | 1x | 0.65x – 0.75x |
| Pauses | Normal | Extended (2–3x longer between sentences) |
| Commentary | Included | Off by default |
| Screen | Normal | Black screen, zero UI after 5 seconds |
| Timer | None | Sleep timer (15 / 30 / 45 / 60 min) with fade-out |
| Resume | Normal | "You fell asleep at 14:22 — resume from here?" |
When the sleep timer ends, Strata bookmarks the exact timestamp. Next session: "You fell asleep around here last night — want to back up 5 minutes?"
- 😴 Sleep-Rated: Meditations, Confessions, poetry, nature writing, gentle philosophy
⚠️ Not Sleep-Rated: War narratives, dense economics, high-conflict fiction
| Tier | Price | Access |
|---|---|---|
| Free | $0 | First 3 episodes of each book (free preview); YouTube + podcast RSS |
| Pro | $5.99/mo or $39.99/yr | Full library, all app features, no ads |
The authoritative price is whatever STRIPE_PRO_PRICE_ID points to in the
Stripe configuration; the figures above are the product target. The free/paid
boundary is enforced server-side by the audio stream endpoint (the first
FREE_PREVIEW_COUNT = 3 episodes of a book are free), and entitlement is read
from the database, not the JWT, so upgrades take effect immediately.
Description: Free listeners get one episode per week on the public feed. Subscribers get the full library on-demand immediately.
Pro Conversion Triggers:
- User wants to binge past the weekly episode
- User wants voice selection, abridgment, notes, exports
- User submits a book for processing
- User wants Atmospheres track control
| Stream | Description |
|---|---|
| Pro Subscriptions | $5.99/mo or $39.99/yr |
| YouTube AdSense | Free public episode feed generates ad revenue |
| Podcast Sponsorships | Aligned sponsors (Catholic media, education, self-improvement) |
| Book Request Processing | Optional one-time fee for priority book processing |
| Voice Pack Bundles | Curated premium voice packs |
| Future: Licensing | License pipeline to publishers, seminaries, homeschool programs |
- Full automated content pipeline (ingest URL → segment → tag → commentary → render → assemble)
- Basic episode player with commentary toggle
- 3 ElevenLabs voice options
- Pre-mixed atmosphere audio (Free)
- Dual-track atmosphere control (Pro)
- Sleep Mode
- YouTube channel + podcast RSS live
- Freemium subscription (Stripe)
- Book Request Queue (user-facing pipeline UI — Phase 2)
- Voice clone
- Notion/Obsidian export
- Bible / New Testament series
- Full social/community features
The pipeline IS the product. Rather than manually ingesting Brothers Karamazov once, the first 4 hours of work should build the automated pipeline that processes any Gutenberg URL end-to-end. This investment pays off immediately — processing one book manually takes the same time as building the automation that processes every future book in minutes.
This pipeline also becomes the user-facing Book Request feature in Phase 2: Pro users paste a Gutenberg URL, the pipeline runs automatically, and the processed book appears in the catalog within 48–72 hours. The internal tool IS the future product feature.
Week 1 pipeline target: Paste https://gutenberg.org/ebooks/28054 → fully assembled episodes with narration + commentary + atmospheres → ready for player.
| Week | Focus | Deliverables |
|---|---|---|
| 1 | Pipeline | Gutenberg ingest → Claude segment + tag + commentary → ElevenLabs render → pydub assembly. End-to-end automated for any URL. |
| 2 | Render & Polish | Multi-voice characters, atmosphere tracks, dual-file output (pre-mixed free + separate Pro tracks) |
| 3 | Player & App | Basic Next.js player, dual-track audio, Sleep Mode, commentary toggle |
| 4 | Launch | YouTube channel live, podcast RSS live, Stripe subscription |
| 5 | Polish | Voice selection, notes/highlights, abridgment toggle |
| 6 | Engagement | AI Chat (book-specific RAG), Must Read tags, shareable quote cards |
| 7 | Expansion | Run America 250 pipeline (Common Sense, Federalist Papers) |
| 8 | Growth | TikTok AI scenes campaign, podcast guesting, community outreach |
| Risk | Mitigation |
|---|---|
| AI voice quality | ElevenLabs multi-voice, user voice choice, Sleep Mode optimization |
| ChatGPT competition | Trained-on-book RAG outperforms generic chat; UX is passive listening not active chat |
| Free tier too generous | Tenderfoot model: free = 1/week, Pro = full library now |
| 8-week timeline | Pipeline first, ruthlessly cut scope, Brothers Karamazov only for MVP |
| Copyright | Public domain only; ElevenLabs output is royalty-free |
| Theological accuracy | Review layer before NT release; commentary flagged as interpretive |
- YouTube first: Channel as top-of-funnel; each video is a free episode pointing to the app
- Podcast: Auto-distribute via Transistor.fm to Spotify, Apple Podcasts, Amazon Music
- Catholic / Christian communities: Hallow audience, Catholic Reddit, RCIA groups
- Homeschool community: Co-ops, Facebook groups, curriculum forums
- ADHD community: Reddit, newsletters, influencer partnerships
Strata should not launch with a random library. Initial catalog should be built around cultural moments that already have search demand, press momentum, and community energy.
| Title | Priority |
|---|---|
| Common Sense — Thomas Paine | Tier 1 |
| Federalist Papers — Hamilton, Madison, Jay | Tier 1 |
| Declaration of Independence + Constitution | Tier 1 |
| Democracy in America — Tocqueville | Tier 1 |
| The Wealth of Nations — Adam Smith | Tier 2 |
| Poor Richard's Almanack — Franklin | Tier 2 |
| Founders' Letters | Tier 2 |
| Two Treatises of Government — Locke | Tier 3 |
| Category | Works |
|---|---|
| Church Fathers | Confessions, City of God, On the Incarnation, Justin Martyr, Tertullian |
| Medieval Doctors | Summa Theologica selections, Proslogion |
| Mystics | Imitation of Christ, Interior Castle, Dark Night of the Soul, Story of a Soul |
| Apologists | Orthodoxy, The Everlasting Man, Newman works |
| Saints' Lives | Butler's Lives of the Saints |
- The Odyssey (Christopher Nolan, Zendaya, Tom Holland)
- The Count of Monte Cristo
- Pride and Prejudice
- Lord of the Flies
- The Aeneid
- Brothers Karamazov (MVP — immediate launch, Hallow momentum)
- America 250 series (July 4, 2026)
- Catholic Fathers and Mystics (sustained niche growth)
- Stoics and Great Books (evergreen expansion)
| Week | Theme | Primary Release |
|---|---|---|
| Week 1 | Founding Primer | Common Sense — Thomas Paine |
| Week 2 | Founding Documents | Declaration of Independence + Constitution |
| Week 3 | Federalist Start | Federalist Papers Part 1 |
| Week 4 | Founding Letters | Selected Founders' Letters |
| Week 5 | American Character | Democracy in America Part 1 |
| Week 6 | Economic Foundation | The Wealth of Nations Part 1 |
| Week 7 | Federalist Continuation | Federalist Papers Part 2 |
| Week 8 | Civic Finish | Poor Richard's Almanack |
| Week 9 | Catholic Launch 1 | Confessions Part 1 |
| Week 10 | Catholic Launch 2 | On the Incarnation |
| Week 11 | Stoic Reset | Meditations Part 1 |
| Week 12 | Prestige Close | The Republic Part 1 |
| Tier | Description | Use Case |
|---|---|---|
| Tier 1 — AI Launch Edition | AI narration, AI commentary, AI atmospheres, pre-mixed | Default launch version |
| Tier 2 — Enhanced AI Edition | Better voice selection, separate tracks, more polish | Pro users and recurring titles |
| Tier 3 — Human Read Edition | Community-sourced or commissioned human narration | Top-performing books, prestige releases |
Objective: Generate high-quality AI movie scenes from classic books to drive awareness and app downloads.
| Format | Example | Tools |
|---|---|---|
| 30-sec scene recreation | "Grand Inquisitor scene from Brothers Karamazov" | Luma Dream Machine + ElevenLabs |
| Quote + visual | Chesterton quote with cinematic background | Runway + text overlay |
| "What if" adaptation | "Federalist Papers as a modern movie trailer" | Higgsfield AI + cinematic style |
| Sleep Mode teaser | "Augustine's Confessions with Gregorian chant" | Lo-fi aesthetic + ambient sound |
Legal risk: Extremely low. Public domain source material + AI-generated output = no copyright exposure.
Production flow:
- Claude generates scene description from book passage
- Luma Dream Machine / Runway generates cinematic clip
- ElevenLabs voices the narration / dialogue
- CapCut / Premiere Rush adds text, music, CTA
- Post to TikTok, Reels, YouTube Shorts simultaneously
Cadence: 1 book per month or every other month for theological review.
| Priority | Book | Rationale |
|---|---|---|
| 1 | Gospel of Mark | Shortest, most accessible |
| 2 | Gospel of John | Theological depth |
| 3 | Acts | Narrative momentum |
| 4 | Romans | Foundational theology |
| 5 | 1 Corinthians | Practical application |
| 6+ | Epistles | Group by theme |
Timeline: Background build starting immediately, first releases Q4 2026 or Q1 2027.
| Risk | Likelihood | Mitigation |
|---|---|---|
| Copyright infringement | None | Source material is public domain |
| AI output copyright | None | AI-generated works not copyrightable (SCOTUS 2026) |
| Trademark | Very low | Use scene descriptions, not trademarked names |
| Platform TOS violations | Low | Luma/Runway allow commercial use |
| YouTube Content ID | Very low | Public domain source = no Content ID match |
The single most important architectural decision in Strata's build is this: the internal book processing pipeline should be built as a real application from day one, not as a one-time manual script.
Why this matters:
The temptation is to manually process Brothers Karamazov — copy the text, paste it into Claude, manually run ElevenLabs, manually stitch audio files. This takes ~4 hours and produces one book.
The smarter investment is to spend those same 4 hours building the automated pipeline (n8n + Claude + ElevenLabs + pydub + Supabase), then paste in the Gutenberg URL and let it run. The pipeline processes Brothers Karamazov automatically AND every future book automatically. The pipeline gets debugged on a real book with real edge cases.
This pipeline becomes the user-facing product: In Phase 2, when Pro users submit a book request, this is the exact same pipeline that runs — it just gets triggered by a user form submission instead of an internal admin action. The internal tool IS the future feature.
Week 1 target: gutenberg.org/ebooks/28054 in → fully assembled, tagged, narrated, atmosphere-mixed episodes out.
What is Content Curator?
Content Curator (see: github.com/bitskc/content-curator) is a companion product that monitors YouTube channels, podcast RSS feeds, and blogs — automatically surfacing content that a user's audience should know about and formatting it for social posting. It is an intelligent feed monitor + social publishing assistant, not a manual snipping tool.
Strata and Content Curator share the same core pipeline architecture:
| Pipeline Step | Strata | Content Curator |
|---|---|---|
| Ingest source | Gutenberg URL | YouTube / Podcast RSS URL |
| Pull text | Book text | Transcript via Whisper / RSS |
| Segment content | Claude splits into chapters | Claude splits into episode chunks |
| AI analysis | Tag mood, themes, must-read | Score shareability, match to audience |
| AI output | Commentary + atmosphere prompts | Social captions + platform formatting |
| Storage | Supabase | Supabase |
| Distribution | YouTube + podcast + app | Social media queue |
Approximately 70% of Strata's pipeline code is directly reusable for Content Curator. The Claude prompts, n8n workflow structure, Supabase schema patterns, and ElevenLabs integrations are all shared or near-identical.
Strata feeds Content Curator directly:
Strata publishes a weekly podcast and YouTube series. A Content Curator user who follows philosophy, theology, or Catholic content would add Strata's RSS as a monitored source. CC automatically surfaces the most shareable quotes from each Strata episode and formats them for LinkedIn or X. Strata generates high-quality source material that Content Curator's pipeline then processes — the two products are naturally complementary.
Build sequencing recommendation:
| Phase | Product | Rationale |
|---|---|---|
| Phase 1 (Now) | Strata MVP | Public domain moat, no legal risk, builds the pipeline |
| Phase 2 (Month 3–6) | Strata catalog expansion | Prove pipeline on 10+ books, grow audience |
| Phase 3 (Month 6–9) | Content Curator MVP | Reuse ~70% of Strata pipeline, add social scheduling layer |
| Phase 4 (Month 9+) | Shared platform layer | Unified auth, shared subscription, Strata as CC source |
Content Curator is Strata's B2B counterpart:
- Strata = B2C consumer product ($5.99/mo), high volume
- Content Curator = B2B prosumer product ($29–59/mo), lower volume, higher LTV
Building Strata first subsidizes Content Curator's development. By the time CC is ready to build, the hardest infrastructure work is already paid for.
Market Context: Hallow's 2026 Lent Pray40 challenge (1.6M+ participants) featured The Brothers Karamazov as its anchor book. Hallow provided surface-level treatment. Strata delivers the full guided deep dive.
Public Domain Status: Fully public domain in the U.S. (1879-1880). Available via Project Gutenberg.
Launch Recommendation:
- Position: "The deeper dive Hallow participants are looking for"
- Abridgment profile: "Theological Core" — Alyosha / Ivan / Grand Inquisitor focus
- Atmosphere: Cathedral reverb, contemplative piano, soft Gregorian chant
- Must Read flags: Grand Inquisitor chapter, Alyosha's key conversations
Strata PRD v2.1 — April 2026