Knowledge Exporter

Turning proprietary SaaS knowledge into a portable, diffable corpus

A provider-abstracted export pipeline that converts Freshdesk Solutions and Confluence content into deterministic Markdown, using SHA-256 comparison so reruns update only what actually changed.

Freshdesk Exporter scope-selection step showing four categories with folder and article counts, and an API Reference category peek listing REST Endpoints and Webhooks & Events folders with their article counts. — Source hierarchy and article counts stay visible before export, making scope an explicit decision rather than an all-or-nothing sync.

Role: Developer Support Engineer
Stack: Next.js • TypeScript • Freshdesk • Confluence • Markdown • Vitest

The Problem

Support and internal documentation accumulate in SaaS platforms as proprietary HTML — hard to audit, version, migrate, or reuse outside the platform that generated it. Freshdesk and Confluence each hold that knowledge in a different shape, with no built-in path to a portable copy.

The goal wasn't to replace either platform. It was a local, versionable copy of the knowledge inside them — one a rerun could trust, not just repeat.

Two Sources, One Shared Pipeline

Freshdesk and Confluence don't share a data model. There's no AI or LLM anywhere in this pipeline, either — every transformation is deterministic, so the same source content always produces the same Markdown output.

Freshdesk: Category → Folder → Article

A fixed three-level hierarchy. Scope is selected by category, and the folders and articles beneath it export together.

Confluence: Nested Page Tree

Pages nest to arbitrary depth inside a space, mirrored directly in Explore's tree view rather than flattened into fixed levels.

Shared Conversion Utilities

Both providers route through the same HTML-to-Markdown conversion, content-quality cleanup, and stable path/slug generation — but each keeps its own source-traversal logic; there's no single normalized document schema both funnel through first.

Server-Side Credentials Only

API credentials for both providers are handled server-side. Nothing touches the browser.

Confluence Explore view of the Engineering space, showing a nested page tree with Webhook Delivery Service expanded into Retry Policy and Signing & Verification child pages, and the Signing & Verification page rendered on the right. — Explore mirrors Confluence's actual nested hierarchy — though today's export writes those pages flatter inside the space folder rather than preserving the full tree on disk.

Every Rerun Is a Diff, Not a Rewrite

Every export run re-fetches the selected source content from Freshdesk or Confluence — nothing about the fetch itself is incremental. Idempotency lives entirely in the write step: each converted document is content-hashed with SHA-256 and compared against what's already on disk, so unchanged output is skipped rather than rewritten, and only real content changes get written as a create or an update.

Hashing decides what changed. Nothing else does.

The result is a per-run change report — created, updated, skipped, and failed counts, plus the exact list of touched paths — so a rerun is something you can audit, not just trust.

Export File Changes dialog showing 3 changed files (1 created, 2 updated), 42 skipped, and 0 failed, with the changed files listed by path and size. — SHA-256 comparison turns a rerun into an actual diff: unchanged files skip cleanly, and only real content changes show up as created or updated.

A Portable, Inspectable Corpus

The export doesn't stay trapped inside a pipeline or a source SaaS platform. Each run writes real files to a local export root — Markdown documents alongside a SUMMARY.md index and a report.json manifest — so the corpus is something you can open, grep, diff, or hand to another tool directly. Provenance for a given run lives in that manifest rather than scattered across individual file frontmatter, keeping the Markdown itself clean and portable.

Exports inventory for the freshdesk export root showing 12 folders, 47 files, 46 Markdown documents, and 51.5 KB total size, alongside a file tree and a rendered Markdown preview of 'Verifying Webhook Signatures.' — The finished export is a browsable local Markdown corpus — structured files and rendered content stay inspectable long after the source API call is gone.

A separate tool, Verbatim, can consume Markdown like this as retrievable source material — the two are complementary, not integrated: there's no shared database, live sync, or direct handoff between them today.

Results

Two structurally different SaaS sources — Freshdesk and Confluence — export through one deterministic pipeline into portable Markdown.
Repeat runs skip unchanged content and touch only what actually changed, instead of rewriting the entire export.
Every run produces an auditable change report: created, updated, skipped, and failed counts, with the exact paths touched.
Scope, run changes, and the finished export are all inspectable — nothing happens silently.
The finished export is a real local filesystem artifact: browsable, greppable, and usable by other tools without depending on the source platform.
API credentials for both providers stay server-side; nothing touches the browser.

Technical Summary

Application: Next.js • TypeScript • Vitest
Conversion: HTML → Markdown (Turndown) • Content-quality normalization • Stable path/slug generation • Oversized-content splitting
Providers: Freshdesk Solutions API (paginated) • Confluence Cloud API (paginated) • Server-side credentials
Reliability: SHA-256 idempotent writes • Per-run change reports • Live export progress

What I'd Build Next

Preserve full source hierarchy on export, not just in ExploreRetry/backoff and resumable job state for long-running exportsIncremental source sync, instead of re-fetching on every runAttachment downloading and inter-document link rewriting

Let’s Work Together

Interested in working together or have a question? Reach out and let’s talk!

Contact Me Connect on LinkedIn