Last updated

Overview Page Restructure: Graph-Aware Focus

User Feedback

  • "7 products" messaging doesn't make sense - that's not a value prop, it's just counting
  • Need to focus on ONE concept: turning data to intelligence with graph-aware embeddings
  • This should be THE main thing, then mention "we also have other things"
  • Remove homepage (redirect to /overview)
  • Remove pricing page
  • Update nav: add "Dashboard Login" and "Products" links
  • Logo should link to papr.ai/landing

Core Insight

Graph-aware embeddings is the differentiation. Everything else (vector storage, chat memory, document extraction, model proxy) is supporting infrastructure that enables the main value prop.

New Messaging Structure

1. Hero / Opening (The Big Idea)

Headline: Turn Data Into Intelligence

Concept: Traditional vector search treats everything the same—a bug report ranks equal to a feature request, yesterday's code equal to last year's. Graph-aware embeddings transform your vector space to represent a structured graph of meaning: temporal context, topical relationships, domain-specific dimensions.

Result: Not just semantic similarity, but domain-aware precision. Code search that understands programming language and API patterns. Scientific search that maps methodology and evidence strength. Custom schemas for your specific domain.

Proof Point: Ranked #1 on Stanford's STaRK benchmark (91%+ accuracy)

2. How It Works (The Technical Beat)

Three-step transformation:

  1. Structured extraction - 14 frequency bands encode domain dimensions (language, intent, entities, temporal context)
  2. Graph-aware encoding - Vector space reshaped to represent relationships, not just similarity
  3. Domain-tuned retrieval - Built-in schemas (code, science) or register your own

Show a before/after example:

  • Before: Search "authentication" → Returns everything with "auth" keyword
  • After: Search "authentication" → Returns JWT implementation in Python with security best practices (filtered by language, ranked by relevance to auth pattern)

3. Built-in Domains (Proof It Works)

  • Code Search (cosqa): +5.5% NDCG@10 over cross-encoder baseline
  • Scientific Claims (scifact): +36% NDCG@10 over baseline
  • General Purpose: Domain-agnostic starting point

Link to detailed guide: Graph-aware embeddings →

4. The Infrastructure (Also Included)

"To make this work, we built a complete memory platform:"

  • Vector Memory - Store and search any content
  • Chat Memory - Conversation storage with compression
  • Document Intelligence - Extract structure from PDFs
  • Knowledge Graphs - Map entity relationships
  • AI Model Proxy - Unified multi-model API
  • Sync & Portability - Local/cloud sync

Each product can be used standalone or together. See all products →

5. Start Building (Call to Action)

  • Quick start (5 min)
  • Try graph-aware search with built-in schemas
  • Design a custom schema for your domain
  • Explore other products

redocly.yaml updates:

logo:
  link: https://papr.ai/landing  # External link to marketing site

navbar:
  items:
    - page: overview/index.md
      label: Docs
    - page: overview/products.md
      label: Products
    - href: https://dashboard.papr.ai
      label: Dashboard Login

Remove pricing page entirely (delete from nav).

Homepage redirect

Option 1: Delete index.page.tsx and make overview/index.md the root page Option 2: Keep index.page.tsx but make it redirect immediately to /overview

Recommend Option 1: Just don't have a separate homepage - docs ARE the homepage.

Decision Tree

Should we update the decision tree to also focus on graph-aware as the starting point, with other products as optional add-ons?

Current decision tree starts with use case (chat vs docs vs generic). New decision tree could start with: "Do you need domain-aware search or just semantic similarity?"

Implementation Checklist

  1. ✅ Create this planning doc
  2. ⬜ Update redocly.yaml (logo link, navbar)
  3. ⬜ Delete or hide pricing.page.tsx
  4. ⬜ Rewrite overview/index.md with graph-aware focus
  5. ⬜ Ensure overview/products.md still exists for detailed product breakdown
  6. ⬜ Test navigation flow
  7. ⬜ Validate links and structure

Key Messaging Principles

  • One hero: Graph-aware embeddings as the transformation
  • Concrete examples: Before/after search results
  • Built-in proof: STaRK benchmark, built-in schemas with performance gains
  • Infrastructure as enabler: All other products support the main value prop
  • Clear CTAs: Quick start, explore schemas, design custom domains