Sina Bolouri

Project

ReasonLoom

ReasonLoom turns permitted tutorial transcripts into source-aware educational article drafts for human review and draft-only publishing to Monograph.

deployedNext.jsReactTypeScriptTailwind CSSAuth.jsDrizzle ORMPostgreSQLInngestVercel
Visit Project

ReasonLoom

ReasonLoom is a private educational publishing workspace that transforms permitted tutorial transcripts into structured, source-aware article drafts. It matters because spoken tutorials often contain useful explanations and reasoning that are difficult to search, review, and preserve; ReasonLoom turns that material into editable learning content while keeping human review and publication control explicit.

Problem

Tutorial transcripts are usually optimized for speech, not durable reading. They contain repeated captions, filler, malformed line breaks, missing visual context, and reasoning spread across long passages. A generic summary can also flatten important details, introduce unsupported claims, or lose the connection between an article and its source.

The publishing workflow adds a second concern: private source material should not become public automatically, and the writing system should not share a database with the public publishing product.

Solution

ReasonLoom provides a transcript-to-article workflow with:

  • Explicit permission confirmation before generation.
  • TXT, Markdown, SRT, and VTT parsing.
  • Deterministic cleaning, warning extraction, token estimation, and topic-aware segmentation.
  • Segment analysis and concept-graph synthesis behind a provider-neutral AI interface.
  • Educational article outlines and sections with support labels, validation findings, and review warnings.
  • Editable Markdown articles with immutable version snapshots and version restore.
  • Prerequisite and related-concept records for knowledge organization.
  • Draft export and a signed Monograph connector that never requests automatic public publication.
  • Auth.js sessions for the first-party workspace and scoped API-key authentication for server clients.

The current deterministic mock provider supports local development, tests, and demonstrations without paid model calls. Gemini, OpenRouter, and Ollama adapters exist at the interface boundary, but their implementations currently reuse the mock behavior rather than making production model requests.

User Experience

An authorized administrator can paste or upload a transcript, provide source metadata, confirm permission, preview preprocessing, choose article settings, and start generation. The workspace shows named processing stages before presenting an article for review.

The administrator can edit Markdown, inspect warnings and validation status, regenerate content, inspect version history, restore versions, manage prerequisites and relationships, export a draft, and send a reviewed draft to Monograph when the article is verified or ready to publish. Public pages provide product, privacy, terms, and API documentation; workspace routes are protected.

Current Status

ReasonLoom is marked deployed based on the project owner's deployment at https://reasonloom.sinabolouri.com. The Next.js application, deterministic mock pipeline, workspace, API surface, database schema, documentation, and automated tests are implemented.

Production integration is not fully verified from the available environment. The live hostname did not resolve during this audit on 2026-09-17, and the active repository factory still selects the local JSON store. Real Gemini, OpenRouter, and Ollama calls, Neon persistence, Inngest execution, OAuth credentials, and the Monograph receiving endpoint require configuration or implementation verification.

Tech Stack

  • Application: Next.js App Router, React, TypeScript strict mode.
  • UI: Tailwind CSS, Radix Slot, Lucide icons, CodeMirror, React Markdown, Shiki.
  • Authentication: Auth.js with Google and GitHub configuration hooks, an administrator email allowlist, and development-only auth.
  • Data model: Drizzle ORM schema and PostgreSQL migration, with Neon as the documented production target.
  • Active development storage: local JSON repository.
  • Background jobs: Inngest client, route, and generation function wiring; local generation runs inline.
  • AI abstraction: deterministic mock provider plus Gemini, OpenRouter, and Ollama adapter boundaries.
  • Validation: Zod schemas and deterministic/article-provider validation stages.
  • Testing: Vitest, Testing Library, and Playwright.
  • Deployment direction: Vercel, Neon, and Inngest.