Skip to content

overview.md

Latest commit

 

History

History
203 lines (160 loc) · 6.83 KB

overview.md

File metadata and controls

203 lines (160 loc) · 6.83 KB

BioLight × Open Notebook — Overview (North Star)

TL;DR

We are using Open Notebook as the self-hosted “knowledge + scripting + podcast generation” core for BioLight.shop so Don Bailey and Mike Belkowski can produce 3–4 podcast episodes/week with minimal friction.
We will ingest BioLight + Energy Code content, generate show-ready scripts, render audio with their ElevenLabs cloned voices, and output publish-ready assets (MP3 + metadata + show notes).
Future social/content features come later — the first build is the in-house podcast factory.

The Problem We’re Solving

Current workflow is clunky:

  • LM Notebook generates content/audio
  • Export / PDF / manual steps
  • Then run through ElevenLabs
  • Then publish

We want:

  • A single system that holds BioLight knowledge, produces on-brand podcast scripts, renders audio using Mike/Don voice clones, and outputs everything needed to publish — without export gymnastics.

Core Goal (MVP)

Build an internal system that can do:

  1. Ingest BioLight knowledge
  2. Generate an episode (multi-speaker)
  3. Render audio using ElevenLabs voice clones
  4. Export publish-ready package (MP3 + title/description/show notes + links/disclaimer)
  5. (Optional in MVP) Auto-upload/schedule to hosting platform

Actors & Context

  • BioLight.shop: brand + products + blog + podcast pages.
  • Podcast: The Energy Code (Mike hosts; Don frequently co-hosts).
  • Mike Belkowski: owner/host voice + scientific storyteller.
  • Don Bailey: CFO/operator voice + pragmatic translator; wants weekly production to be frictionless.
  • Voices: Mike + Don voice clones already exist in ElevenLabs.

Why Open Notebook is the Base

Open Notebook provides:

  • Content ingestion (links, PDFs, YouTube, txt, etc.)
  • AI notes + summarization + transformations
  • A built-in Podcast Generator system (episodes + speakers + voices)
  • Privacy/self-hosting + API for automation

We will treat Open Notebook as the “brain + factory floor.” We will add a thin “publisher” layer later if needed.

System Concept: “BioLight Podcast Factory”

Input

A producer (usually Don) provides one of:

  • Bullet outline / topic prompt
  • Link dump (BioLight pages + studies + blog posts)
  • A product/topic focus (“mitochondria-first stack”, “methylene blue”, etc.)
  • Guest talking points (optional)

Output

A complete publish-ready bundle:

  • Episode title + alt titles
  • Cold open + segment outline
  • Two-speaker script (Mike + Don)
  • Show notes w/ links
  • Disclaimer block (health/wellness safe)
  • MP3 audio (ElevenLabs voices)
  • Metadata JSON (duration, tags, timestamps if possible)

Data Ingestion Strategy (Knowledge Base)

Create a dedicated notebook:

  • Notebook: BioLight Knowledge Base

Ingest:

  • Product pages (all flagship SKUs + collections)
  • Podcast episode pages (show notes, summaries, timestamps when present)
  • BioLight blog/learn articles
  • Internal docs: product positioning, FAQs, protocols, “do not claim” rules
  • Brand voice rules: tone, words to avoid, disclaimers, audience profile

Key Principle

All generation should cite/anchor to this knowledge base to stay:

  • Accurate
  • On-brand
  • Consistent in claims and language

Voice + Style: Speaker Profiles

We maintain two “Speaker Profiles” (canonical, reused everywhere):

Speaker: Mike

  • Role: host / educator / scientific storyteller
  • Style: confident, curious, structured, myth-busting
  • Output: frames the topic, teaches, makes it memorable

Speaker: Don

  • Role: operator / CFO / practical translator
  • Style: concise, grounded, implementation-focused
  • Output: asks real-world questions, clarifies, adds business/practice perspective

We also define one “Episode Profile”:

  • Format: 2 speakers
  • Structure: Hook → 2–4 teaching blocks → actionable takeaways → soft CTA
  • Guardrails: avoid risky medical claims; always include disclaimer

Guardrails (Non-Negotiable)

BioLight content lives near health claims. We must:

  • Bake in a consistent disclaimer block
  • Maintain a “Do Not Claim” list
  • Prefer phrasing like: “may support”, “emerging evidence suggests”, “talk to your clinician”, etc.
  • Avoid diagnosis/treatment promises
  • Avoid implying FDA approval unless explicitly true

These guardrails must be applied:

  • In script generation
  • In show notes generation
  • In any derivative content later

MVP Feature List (Build Order)

Phase A — Platform + Knowledge

  • Self-host Open Notebook
  • Secure it (auth + keys)
  • Create BioLight Knowledge Base notebook
  • Ingest initial sources (products + recent podcast episodes + top blog content)

Phase B — Episode Generator

  • Create Mike + Don Speaker Profiles
  • Create “Energy Code” Episode Profile template
  • Implement transformations:
    • Episode titles
    • Outline
    • Full script
    • Show notes + links
    • Disclaimer block (always)

Phase C — Audio Rendering

  • Use ElevenLabs API with:
    • Mike voice ID
    • Don voice ID
  • Render script → MP3
  • Output MP3 + metadata bundle

Phase D (Optional in MVP) — Publish Assist

  • Export a ready-to-upload package
  • (Later) Direct upload/scheduling integration to hosting platform

What Success Looks Like (MVP)

Don can produce an episode with:

  • < 10 minutes of input work
  • Mostly one-click automation after providing topic/links
  • Consistent Mike/Don voice + format
  • Publish-ready outputs with minimal edits

Target throughput:

  • 3–4 episodes per week reliably

Architectural Notes

  • Open Notebook is the “system of record” for knowledge and generation.
  • We automate via Open Notebook’s API (create sources, trigger transforms, trigger podcast generation, poll job status).
  • ElevenLabs is only the rendering engine for the voices (TTS output).
  • Publishing is a separate module; don’t block MVP on it.

Future Extensions (Not MVP)

We intentionally defer:

  • Social media clip generation (shorts, reels, TikTok scripts)
  • Auto-generated carousels + newsletters
  • Auto-cut video avatars / facecam compositing
  • Full distribution automation across platforms

Those will be layered on after the podcast factory is stable.

Open Questions (Track, Don’t Block MVP)

  • Where do we host the final MP3s in v1 (manual upload vs API)?
  • Do we need per-episode approvals/versioning?
  • Do we need transcripts auto-published on BioLight.shop?
  • How strict should topic gating be (avoid sensitive medical claims)?

Immediate Next Steps

  1. Stand up Open Notebook locally (docker compose).
  2. Create the BioLight Knowledge Base notebook.
  3. Ingest: top products + last ~20 podcast pages + highest-performing blog posts.
  4. Implement speaker profiles + “Energy Code” episode profile.
  5. Wire ElevenLabs voice IDs + generate first internal test episode.
  6. Iterate until output needs minimal edits.