← Back to Blog
December 5, 2024

60 Docs in 60 Minutes: Our Approach to Doc Generation

How do you generate 60 comprehensive documents quickly and consistently? Here's our approach.

The Architecture

Context Ark uses a **kernel-based** architecture:

1. **Intake** — Gather project information through structured Q&A

2. **Kernel** — Compile answers into a structured JSON "brain"

3. **Generation** — Use the kernel to generate each document

The kernel is the single source of truth. All documents derive from it.

The Generation Pipeline

We use Inngest for durable background processing:

1. **Batch planning** — Determine which docs to generate

2. **Parallel generation** — Generate docs concurrently

3. **Consistency checks** — Ensure terminology matches

4. **Assembly** — Compile into exportable format

LLM Strategy

We use a tiered model approach:

  • **Budget tier**: DeepSeek, Qwen (fastest, cheapest)
  • **Balanced tier**: Gemini 2.0 Flash (good quality, fast)
  • **Premium tier**: GPT-4o, Claude 3.5 (highest quality)
  • Fallback chains ensure reliability: if one provider fails, we try the next.

    The Result

    60+ documents in under an hour, all consistent and accurate.