AI Companion For Book
Budget: $250 – $750 USD
RAG index search engine for my book.
Build a web-based companion to my book The Carbonated Body that helps readers find where concepts are discussed (chapters/sections/page numbers) without replacing reading. The AI must behave like a navigator / librarian, not a summarizer or “answer engine.” It will act like a dynamic index for the book.
The system should:
Point readers to the exact places in the book where answers are explained
Encourage sequential reading
Provide only minimal “teaser” context
Avoid detailed explanations or full summaries that let users skip the book
Core Concept: “Locate, Don’t Explain”
The AI is allowed to:
Identify relevant chapters/sections
Return page numbers (or Kindle location equivalents if needed)
Quote very short snippets (optional; strict limit)
Provide a 1–2 sentence “why this is relevant” teaser
Recommend reading order when concepts depend on earlier chapters
The AI is NOT allowed to:
Fully answer conceptual questions
Provide multi-point summaries of mechanisms
Provide “here are the steps” style content that replaces a chapter
Output large excerpts of the book
User Experience Requirements
Primary UI: Search + Results
A search box where users can type:
Keywords (“Bohr effect”, “microclots”)
Natural questions (“How does CO₂ affect inflammation?”)
Use cases (“What should I read if I have POTS?”)
The response format should be consistent:
1) Best places to read (ranked)
Chapter title
Section/subsection title
Page range
Confidence score (optional)
2) Recommended reading path
“Start here” → “then read” → “then read”
The AI should guide sequence when the idea builds across chapters
3) Minimal context
Max 1–2 short sentences explaining why those sections match
No deeper explanation
4) Optional short quote (strictly controlled)
Optional snippet, max ~1–2 sentences, and a hard character limit (e.g., 250–400 chars)
Never multiple paragraphs
5) “Open / Jump to”
Links to a preview-like reference (not full text) that shows:
Chapter/section name
Page numbers
Short excerpt or “page image” (depending on rights/format)
“Reading-First” Guardrails (Hard Rules)
1) Refuse detailed answers
If a user asks “Explain X”, the AI should respond with:
Where to read it
A reading path
A short teaser
A line like: “This is explained in depth in the book sections above.”
2) No mechanism lists
Avoid any response like:
“There are 5 mechanisms…”
“In summary…”
3) Rate limits / friction features (recommended)
Add mild friction to discourage using it as a shortcut:
After N queries per day, prompt: “You’ll get more value by reading Chapter __ next.”
Or require selecting: “Have you read Chapter __ yet?” before showing more references
4) Content protection
Never return more than a small excerpt.
Never allow “give me the full chapter” or “paste the section.”
Detect and refuse requests for large portions of text.
Data / Content Ingestion
Source material
I will provide:
The final PDF manuscript (or EPUB / DOCX)
A table of contents with chapter/section headings
Page numbers for print layout (if available)
Indexing approach (recommended)
Use RAG:
Chunk the book by semantic sections (preferably aligned to headings)
Store embeddings with metadata:
Chapter number
Chapter title
Section title
Page start/end
Keywords/tags (optional)
Page mapping
Important: The AI must return accurate page ranges.
If PDF is the canonical source: map chunk → PDF page numbers.
If print layout differs, we’ll need a stable mapping table.
Build a web-based companion to my book The Carbonated Body that helps readers find where concepts are discussed (chapters/sections/page numbers) without replacing reading. The AI must behave like a navigator / librarian, not a summarizer or “answer engine.” It will act like a dynamic index for the book.
The system should:
Point readers to the exact places in the book where answers are explained
Encourage sequential reading
Provide only minimal “teaser” context
Avoid detailed explanations or full summaries that let users skip the book
Core Concept: “Locate, Don’t Explain”
The AI is allowed to:
Identify relevant chapters/sections
Return page numbers (or Kindle location equivalents if needed)
Quote very short snippets (optional; strict limit)
Provide a 1–2 sentence “why this is relevant” teaser
Recommend reading order when concepts depend on earlier chapters
The AI is NOT allowed to:
Fully answer conceptual questions
Provide multi-point summaries of mechanisms
Provide “here are the steps” style content that replaces a chapter
Output large excerpts of the book
User Experience Requirements
Primary UI: Search + Results
A search box where users can type:
Keywords (“Bohr effect”, “microclots”)
Natural questions (“How does CO₂ affect inflammation?”)
Use cases (“What should I read if I have POTS?”)
The response format should be consistent:
1) Best places to read (ranked)
Chapter title
Section/subsection title
Page range
Confidence score (optional)
2) Recommended reading path
“Start here” → “then read” → “then read”
The AI should guide sequence when the idea builds across chapters
3) Minimal context
Max 1–2 short sentences explaining why those sections match
No deeper explanation
4) Optional short quote (strictly controlled)
Optional snippet, max ~1–2 sentences, and a hard character limit (e.g., 250–400 chars)
Never multiple paragraphs
5) “Open / Jump to”
Links to a preview-like reference (not full text) that shows:
Chapter/section name
Page numbers
Short excerpt or “page image” (depending on rights/format)
“Reading-First” Guardrails (Hard Rules)
1) Refuse detailed answers
If a user asks “Explain X”, the AI should respond with:
Where to read it
A reading path
A short teaser
A line like: “This is explained in depth in the book sections above.”
2) No mechanism lists
Avoid any response like:
“There are 5 mechanisms…”
“In summary…”
3) Rate limits / friction features (recommended)
Add mild friction to discourage using it as a shortcut:
After N queries per day, prompt: “You’ll get more value by reading Chapter __ next.”
Or require selecting: “Have you read Chapter __ yet?” before showing more references
4) Content protection
Never return more than a small excerpt.
Never allow “give me the full chapter” or “paste the section.”
Detect and refuse requests for large portions of text.
Data / Content Ingestion
Source material
I will provide:
The final PDF manuscript (or EPUB / DOCX)
A table of contents with chapter/section headings
Page numbers for print layout (if available)
Indexing approach (recommended)
Use RAG:
Chunk the book by semantic sections (preferably aligned to headings)
Store embeddings with metadata:
Chapter number
Chapter title
Section title
Page start/end
Keywords/tags (optional)
Page mapping
Important: The AI must return accurate page ranges.
If PDF is the canonical source: map chunk → PDF page numbers.
If print layout differs, we’ll need a stable mapping table.