Book recommender

A mini project to help me decide what to read next. The idea: score all my unread books against my interests and garden context, then let situational vibe chips re-rank the results based on what I actually need right now.

Why

I have stacks of unread books, both physical and on my e-reader. Picking the next one is surprisingly hard. I don’t want algorithmic recommendations based on what other people read. I want something that knows me: my library, my current interests, my energy level.

Design constraints

• No AI: pure logic and scoring. How far can structured matching get us?
• Free: no API keys, no subscriptions
• Visual: an interactive HTML dashboard (single file, like a vibe-coded prototype)
• Personal: reads directly from my digital garden and library data
• Integrated: reads from and writes back to the garden library. The library page is the single source of truth

How it works

Input:

• The garden library (all books with audit labels: status, type, purpose, ratings)
• Vibe chip selection (optional, client-side)

Context (pulled automatically):

• Garden content tags across all collections
• Active garden projects (tagged project)

Scoring dimensions (Python, run once):

1. Topic match (30%): keyword overlap with three interest threads: systems/ecology, language/storytelling, philosophy/depth
2. Experience (35%): predicted enjoyment: loved authors, genre density, momentum signals
3. Garden connection (20%): tag and project overlap with active garden work
4. Freshness (15%): rediscovery bonus for forgotten or recently acquired books

Vibe chips (client-side re-ranking):

• Short session: best-scoring book under 250 pages
• Feed the work: re-ranks by garden connection score
• From the pile: re-ranks by freshness score
• Surprise me: widens the shuffle pool to top 15

Output format:

• Not “pick one book” but a reading mix: a lighter book for momentum + a denser book for longer-term reading
• Interactive HTML dashboard with scored book cards

Project log

2026-03-17: kickoff

Researched the approach using Claude. Decided on logic-based scoring over embeddings.

• User research interview (write-up)
• Library audit, 40 books labeled (prototype)
• Library audit follow-up: interview for remaining unlabeled books
• Design the scoring algorithm (design)
• Build the Python script
• Build the HTML dashboard (score breakdowns, pages, reading time, learning layer)
• Add learning layer to dashboard (score tooltips, matched keywords, algorithm explainer)
• Library integration (design): extend library frontmatter with audit fields
• Library integration: update recommender to read from garden library markdown files
• Library integration: write top picks back to library frontmatter
• Add vibe chips: client-side re-ranking with situational chips and a shuffle button
• Library integration: inline audit editing on library page
• Library integration: add “get recommendations” trigger on library page