Scrollabus

Doomscroll your syllabus. Let the algorithm teach you.

Scrollabus is a TikTok-style academic feed. Upload study material — a lecture PDF, notes, or a text paste — and it transforms that content into a personalized stream of short-form educational posts. Six built-in AI study influencer personas each write in a distinct voice and style; users can also create their own custom personas. Students comment, save, interact with posts, and take inline quizzes the same way they would on any creator platform. When a student comments, the persona replies in-character via AI. The feed adapts based on what each user uploads, saves, and engages with. A memory-native agent layer tracks long-term learning patterns and powers an ambient Telegram study companion.

Features

Material-to-feed pipeline: paste text, upload a PDF (with Gemini vision OCR), or link content — Scrollabus generates up to 30 short-form posts per upload across all active personas
Six built-in AI study influencer personas: Lecture Bestie, Exam Gremlin, Problem Grinder, Doodle Prof, Meme Lord, and Study Bard — each with a distinct system prompt, accent color, and teaching style
Custom persona creation: users can author their own study personas (name, emoji, accent color, tone, teaching style, system prompt), make them public, and use them for post and quiz generation
Multimodal posts: each post can be text, image (AI-generated), audio (ElevenLabs TTS narration or Study Bard song), or slideshow (3 Imagen 4 Fast frames + Gemini TTS voiceover, auto-advancing with crossfade) depending on persona and user preferences
Interactive quizzes: multiple-choice, multiple-response, and free-text quiz cards generated from user materials via Gemini 2.5 Flash; each question is voiced in the generating persona's style; auto-graded with AI-powered hint/explanation chat per question (also Gemini 2.5 Flash, Socratic guardrails before answer, full explanation after)
Adaptive feed: ?mode=adaptive biases the feed toward personas the student engages with most, scored by a get_persona_affinity RPC (likes +3, saves +3, dwell ≥ 2 s +1, comments +2)
Dify teaching plan: on every material upload, a Dify workflow analyzes the content and produces a pedagogical plan (priority_personas, emphasis, first_review_concept, quiz_topic); this plan is stored in materials.teaching_plan and passed to n8n to enrich content generation
Reactive content loop (Workflow 7): a scheduled n8n job reads quiz failure hotspots and engagement data, then generates targeted remedial posts (source = 'reactive') via Featherless AI
Persona pulse (Workflow 8): a scheduled n8n job generates community update posts (post_type = 'pulse') in each persona's voice based on aggregate engagement trends
Learner memory layer: HydraDB stores long-term learner profiles, misconceptions, and study history; surfaced across quizzes, Telegram nudges, and the teaching agent
Telegram study companion: students link their Telegram account via a bot deep link; Scrollabus sends proactive study nudges and handles inbound replies (quiz_me, explain_simpler, recap, general questions) via GMI Cloud and learner memory
Dify teaching agent: a configurable Dify workflow orchestrates scheduled follow-ups, quiz delivery, and contextual re-explanation using GMI Cloud (DeepSeek/Llama)
Live AI comment replies: commenting on a post triggers an n8n webhook that calls Featherless AI (DeepSeek-V3.2), which replies in the persona's voice and saves the response to Supabase
Direct messages: users can DM any persona directly, powered by Gemini 2.0 Flash
Explore feed: trending posts from other users' materials, ranked by an engagement score RPC (get_trending_posts); includes live YouTube search via YouTube Data API v3
Community tab: surfaces users who share overlapping interests via get_similar_users RPC
External content ingestion: n8n workflows scrape YouTube (every 6 hours via YouTube Data API v3) and TikTok (public oEmbed) into the external_content table
Creao integration: a scheduled n8n workflow polls a Google Sheet for Creao-generated content every 15 minutes and inserts new posts
Save categories: users organise saved posts into named collections
Persona-filtered feed: filter any feed view by a specific persona slug
PWA manifest: installable as a mobile web app
Row-level security: all tables use Supabase RLS; service-role key is only used inside n8n and Next.js API routes

Tech Stack

Layer	Technology
Framework	Next.js 15 (App Router, TypeScript)
Database & auth	Supabase (Postgres + Auth + Storage)
AI — PDF parsing	Gemini 2.5 Flash (vision), `pdf-parse` (fallback)
AI — DM chat	Gemini 2.0 Flash
AI — comment replies	Featherless AI / DeepSeek-V3.2 (via n8n)
AI — post generation	Featherless AI / DeepSeek (via n8n)
AI — quiz generation	Gemini 2.5 Flash (`lib/quizzes.ts`)
AI — quiz chat	Gemini 2.5 Flash
AI — image generation	Gemini Imagen (via n8n)
AI — slideshow video	Imagen 4 Fast (frames) + Gemini TTS (voiceover) via n8n
AI — video generation	Kling (via n8n, set `KLING_API_KEY` inside n8n)
AI — TTS audio	ElevenLabs (Next.js `/api/tts`), Gemini TTS (via n8n)
AI — teaching agent	Dify workflow + GMI Cloud (DeepSeek/Llama)
AI — Anthropic	Claude SDK (available, not wired to a specific route by default)
Learner memory	HydraDB (`@hydra_db/node`)
Ambient companion	Telegram Bot API — study nudges via Telegram
Automation	n8n (self-hosted, 8 workflows + 5 sub-workflows)
Scheduling & bridge	Creao → Google Sheets → n8n
Social ingestion	YouTube Data API v3, TikTok oEmbed
Styling	Tailwind CSS v3
Animation	Framer Motion
Bottom sheets	Vaul
Deployment	Vercel (assumed)

Architecture

flowchart TD
    User["Browser / PWA"]
    TG["Student Telegram"]

    subgraph Next["Next.js App (App Router)"]
        FeedAPI["/api/feed"]
        CommentsAPI["/api/comments"]
        MaterialsAPI["/api/materials"]
        ParsePDF["/api/materials/parse-pdf"]
        ExploreAPI["/api/explore"]
        CommunityAPI["/api/community"]
        DMAPI["/api/personas/:slug/dm"]
        PersonaAPI["/api/personas/:slug"]
        PersonasAPI["/api/personas"]
        QuizAPI["/api/quizzes + /api/quiz-chat + /api/quiz-responses"]
        TTSAPI["/api/tts"]
        TelegramAPI["/api/telegram/*"]
        PostAPI["/api/posts/:id"]
    end

    subgraph Supabase["Supabase"]
        Auth["Auth"]
        DB["Postgres"]
        Storage["Storage (post-media, avatars)"]
        AffinityRPC["get_persona_affinity RPC"]
        TrendingRPC["get_trending_posts RPC"]
        SimilarRPC["get_similar_users RPC"]
    end

    subgraph N8N["n8n Workflows"]
        W1["Workflow 1: Material → Posts (Orchestrator)"]
        W1a["1a: sub-text"]
        W1b["1b: sub-image"]
        W1c["1c: sub-audio-tts"]
        W1d["1d: sub-audio-bard"]
        W1e["1e: sub-video"]
        W2["Workflow 2: Comment → AI Reply"]
        W3["Workflow 3: Engagement Sync"]
        W4["Workflow 4: Creao Ingest (every 15 min)"]
        W5["Workflow 5: YouTube Fetch (every 6 h)"]
        W6["Workflow 6: TikTok Fetch"]
        W7["Workflow 7: Reactive Content (scheduled)"]
        W8["Workflow 8: Persona Pulse (scheduled)"]
    end

    Gemini["Gemini (PDF parse, quiz gen, quiz chat, DM chat, Imagen, TTS)"]
    Featherless["Featherless AI (DeepSeek-V3.2)"]
    Kling["Kling (video generation via n8n)"]
    ElevenLabs["ElevenLabs (TTS)"]
    HydraDB["HydraDB (learner memory)"]
    Dify["Dify (teaching agent)"]
    GMI["GMI Cloud (DeepSeek/Llama)"]
    TelegramBot["Telegram Bot API"]
    Creao["Creao → Google Sheets"]
    YouTube["YouTube Data API v3"]
    TikTok["TikTok oEmbed"]

    User -->|"HTTPS"| Next
    TG <-->|"Telegram"| TelegramBot
    TelegramBot <-->|"webhook"| TelegramAPI
    TelegramAPI --> HydraDB
    TelegramAPI --> Dify
    TelegramAPI --> GMI
    Dify --> GMI
    Next --> Auth
    Next --> DB
    Next --> Storage
    FeedAPI --> AffinityRPC
    ExploreAPI --> TrendingRPC
    CommunityAPI --> SimilarRPC
    MaterialsAPI -->|"upload text/PDF"| ParsePDF
    ParsePDF -->|"vision OCR"| Gemini
    MaterialsAPI -->|"Dify teaching plan (10 s timeout)"| Dify
    MaterialsAPI -->|"POST webhook (with teaching plan)"| W1
    CommentsAPI -->|"POST webhook"| W2
    DMAPI -->|"chat"| Gemini
    TTSAPI --> ElevenLabs
    QuizAPI --> Gemini
    QuizAPI --> DB

    W1 --> W1a & W1b & W1c & W1d & W1e
    W1a & W1b & W1c & W1d & W1e --> Featherless
    W1b --> Gemini
    W1c --> Gemini
    W1d --> Gemini
    W1e --> Kling
    W1a & W1b & W1c & W1d & W1e -->|"INSERT posts"| DB

    W2 --> Featherless
    W2 -->|"INSERT comment"| DB

    W4 --> Creao
    W4 -->|"INSERT posts"| DB

    W5 --> YouTube
    W5 -->|"INSERT external_content"| DB

    W6 --> TikTok
    W6 -->|"INSERT external_content"| DB

    W7 -->|"quiz failure hotspots"| DB
    W7 --> Featherless
    W7 -->|"INSERT reactive posts"| DB

    W8 -->|"engagement summary"| DB
    W8 --> Featherless
    W8 -->|"INSERT pulse posts"| DB

Data Model

erDiagram
    profiles {
        uuid id PK
        text display_name
        text bio
        text avatar_url
        bool onboarding_completed
        text[] interests
        text[] enabled_personas
        bool enable_av_output
        text telegram_chat_id
        text telegram_username
        bool telegram_enabled
        text telegram_link_token
    }
    personas {
        uuid id PK
        text slug UK
        text name
        text description
        text system_prompt
        text accent_color
        text role_tag
        text emoji
        text tone
        text teaching_style
        uuid created_by FK
        bool is_public
    }
    materials {
        uuid id PK
        uuid user_id FK
        text title
        text raw_text
        text source_type
        jsonb teaching_plan
    }
    posts {
        uuid id PK
        uuid material_id FK
        uuid persona_id FK
        text post_type
        text title
        text body
        text media_type
        text media_url
        text source
        int sort_order
    }
    comments {
        uuid id PK
        uuid post_id FK
        uuid user_id FK
        uuid persona_id FK
        text body
        uuid parent_comment_id FK
    }
    saves {
        uuid user_id FK
        uuid post_id FK
    }
    likes {
        uuid user_id FK
        uuid post_id FK
    }
    save_categories {
        uuid id PK
        uuid user_id FK
        text name
    }
    save_category_memberships {
        uuid save_id FK
        uuid category_id FK
    }
    impressions {
        uuid id PK
        uuid user_id FK
        uuid post_id FK
        int duration_ms
    }
    external_content {
        uuid id PK
        text platform
        text external_id
        text embed_url
        text[] tags
    }
    quizzes {
        uuid id PK
        uuid material_id FK
        text persona_slug
        text question_type
        text question
        jsonb options
        jsonb correct_answer
        text explanation
    }
    quiz_responses {
        uuid id PK
        uuid quiz_id FK
        uuid user_id FK
        jsonb answer
        bool is_correct
    }
    quiz_messages {
        uuid id PK
        uuid quiz_id FK
        uuid user_id FK
        text persona_slug
        text role
        text body
    }
    learner_events {
        uuid id PK
        uuid user_id FK
        text event_type
        jsonb payload
    }

    profiles ||--o{ materials : "uploads"
    profiles ||--o{ saves : "saves"
    profiles ||--o{ likes : "likes"
    profiles ||--o{ impressions : "views"
    profiles ||--o{ comments : "writes"
    profiles ||--o{ save_categories : "owns"
    profiles ||--o{ quiz_responses : "submits"
    profiles ||--o{ quiz_messages : "chats"
    profiles ||--o{ learner_events : "generates"
    profiles ||--o{ personas : "creates (custom)"
    materials ||--o{ posts : "generates"
    materials ||--o{ quizzes : "generates"
    personas ||--o{ posts : "authors"
    posts ||--o{ comments : "receives"
    posts ||--o{ saves : "saved via"
    posts ||--o{ likes : "liked via"
    posts ||--o{ impressions : "tracked via"
    quizzes ||--o{ quiz_responses : "answered via"
    quizzes ||--o{ quiz_messages : "discussed via"
    save_categories ||--o{ save_category_memberships : "contains"

Quickstart

1. Prerequisites

Node.js 20+
A Supabase project
A self-hosted or cloud n8n instance
API keys for the services you want to enable (see Environment Variables)

2. Clone and install

git clone <repo-url> scrollabus
cd scrollabus
npm install

3. Configure environment

cp .env.example .env.local
# Fill in the required values in .env.local

4. Set up Supabase

Run these SQL files against your Supabase project in order:

# 1. Base schema
supabase db push --file supabase/schema.sql

# 2. Seed personas
supabase db push --file supabase/seed.sql

# 3. Incremental migrations (run each in supabase/migrations/ in filename order)

# 4. RPCs
supabase db push --file supabase/engagement-rpc.sql
supabase db push --file supabase/community-rpc.sql
supabase db push --file supabase/external-content-rpc.sql
supabase db push --file supabase/persona-affinity-rpc.sql

5. Import n8n workflows

In n8n, import all JSON files from n8n/ (Workflows 1–8 and sub-workflows 1a–1e).
After importing, open Workflow 1 and replace the placeholder workflowId values in each Execute Workflow node with the actual IDs assigned by your n8n instance.
Set these environment variables inside n8n (not in .env.local):
- FEATHERLESS_API_KEY
- SUPABASE_URL
- SUPABASE_SERVICE_KEY
- KLING_API_KEY (for video generation, Workflow 1e)
- YOUTUBE_API_KEY (for Workflow 5)
- GOOGLE_SHEET_ID (for Workflows 3 & 4)
Activate all workflows.

6. Run

npm run dev
# App runs on http://localhost:3000

How It Works

Material Upload Flow

POST /api/materials
  │
  ├─ INSERT material row (Supabase)
  │
  ├─ writeMemory (HydraDB, fire-and-forget)
  │
  ├─ generateAndSaveQuizzes (Gemini 2.5 Flash, fire-and-forget)
  │   └─ persona-voiced quizzes → INSERT quizzes (Supabase)
  │
  └─ async pipeline:
      ├─ triggerTeachingWorkflow (Dify, 10 s timeout)
      │   └─ returns { emphasis, priority_personas, … }
      │   └─ UPDATE materials.teaching_plan (Supabase)
      │
      └─ triggerMaterialToPost (n8n Webhook, enriched with Dify output)
          └─ Workflow 1 (Orchestrator)
              ├─ 1a: text posts    → Featherless AI → INSERT posts
              ├─ 1b: image posts   → Featherless AI + Imagen → INSERT posts
              ├─ 1c: audio TTS     → Featherless AI + Gemini TTS → INSERT posts
              ├─ 1d: audio bard    → Featherless AI + Gemini TTS → INSERT posts
              └─ 1e: video/slideshow → Kling + Imagen 4 Fast → INSERT posts

Comment Reply Flow

POST /api/comments  →  INSERT comment (Supabase)
                    →  POST n8n webhook (fire-and-forget)
                           └─ Workflow 2: calls Featherless AI (DeepSeek-V3.2)
                                         → INSERT AI reply comment (Supabase)

Quiz Generation

lib/quizzes.ts generates quizzes using Gemini 2.5 Flash with per-persona prompting strategies. Each enabled persona generates QUIZZES_PER_MATERIAL / persona_count questions (minimum 1), all in parallel. Results are bulk-inserted into the quizzes table with a persona_slug column.

Default: QUIZZES_PER_MATERIAL = 3.

Feed Algorithm

The home feed uses exponential-decay slot allocation across the user's most recent five materials (newest → most posts), followed by round-robin interleaving across material buckets so content from different uploads alternates in the scroll.

Decay constant: 0.6 (each older material gets 60% of the slots of the one above it)
Page size: 10 posts (FEED_PAGE_SIZE)
Pagination: per-material cursor map { [materialId]: lastSeenCreatedAt | null } serialized as a JSON string

Adaptive mode (?mode=adaptive): fetches get_persona_affinity scores for the user and reranks posts within each material bucket toward higher-affinity personas. Scoring weights: likes +3, saves +3, dwell ≥ 2000 ms +1, human comments +2.

On fresh feed load (no cursor), the feed route recalls weak/stale learner memories from HydraDB and fires a feed_opened event to the Dify teaching workflow.

Scheduled n8n Workflows

Workflow	Trigger	What it does
4 — Creao Ingest	every 15 min	Polls Google Sheet for Creao posts, inserts into `posts`
5 — YouTube Fetch	every 6 h	Fetches channel videos via YouTube Data API v3, inserts into `external_content`
6 — TikTok Fetch	schedule	Fetches via public oEmbed, inserts into `external_content`
7 — Reactive Content	schedule	Reads quiz failure hotspots + engagement data, generates remedial `reactive` posts via Featherless AI
8 — Persona Pulse	schedule	Reads engagement summaries, generates community `pulse` posts in each persona's voice via Featherless AI

API Routes

Method	Route	Description
`GET`	`/api/feed`	Paginated home feed (cursor-based); supports `?material_id`, `?persona_slug`, `?mode=adaptive`
`GET/POST`	`/api/comments`	List or create a comment; POST triggers n8n Workflow 2 webhook for AI reply
`POST`	`/api/materials`	Upload study material; triggers quiz generation (Gemini 2.5 Flash), Dify teaching plan, and n8n Workflow 1
`DELETE`	`/api/materials/[id]`	Delete a material and its posts
`POST`	`/api/materials/parse-pdf`	Parse a PDF upload: Gemini 2.5 Flash vision (primary) or `pdf-parse` (fallback)
`GET`	`/api/posts/[id]`	Fetch a single post by ID
`GET/POST`	`/api/personas`	List all visible personas; POST creates a custom persona (authenticated user)
`GET/PATCH/DELETE`	`/api/personas/[slug]`	Fetch persona + recent posts; PATCH updates a custom persona (owner only); DELETE removes it
`POST`	`/api/personas/[slug]/dm`	Send a DM to a persona (Gemini 2.0 Flash, stateful chat)
`GET`	`/api/explore`	Trending posts via `get_trending_posts` RPC
`GET`	`/api/explore/external`	External content (YouTube / TikTok)
`GET`	`/api/explore/search`	Live YouTube search via YouTube Data API v3
`GET`	`/api/community`	Similar users via `get_similar_users` RPC
`GET/POST`	`/api/saves`	Save or unsave a post
`GET/POST`	`/api/saved-posts`	Saved posts with optional category filter
`POST`	`/api/saved-posts/memberships`	Assign a saved post to categories
`GET/POST`	`/api/save-categories`	List or create save categories
`GET/POST`	`/api/likes`	Like or unlike a post
`GET`	`/api/likes/posts`	Fetch liked posts
`POST`	`/api/impressions`	Record a feed view impression
`GET/PATCH`	`/api/profile`	Fetch or update profile
`POST`	`/api/profile/avatar`	Upload profile avatar to Supabase Storage
`POST`	`/api/profile/interests`	Update user interests
`GET`	`/api/quizzes`	List quizzes for a material
`POST`	`/api/quizzes/generate`	Generate persona-voiced quizzes from a material (Gemini 2.5 Flash)
`GET/POST`	`/api/quiz-responses`	Fetch or submit a quiz response (auto-graded for MCQ/multiple-response)
`GET/POST`	`/api/quiz-chat`	Fetch or send a message in a quiz's hint/explanation chat (Gemini 2.5 Flash)
`POST`	`/api/tts`	Text-to-speech via ElevenLabs
`GET/POST`	`/api/telegram/connect`	Link Telegram account and toggle study companion
`POST`	`/api/telegram/webhook`	Inbound Telegram message handler (quiz_me, explain_simpler, recap, etc.)
`GET`	`/api/telegram/cron`	Scheduled outbound study nudge dispatcher via Telegram
`POST`	`/api/telegram/setup-webhook`	One-time Telegram webhook registration
`GET`	`/api/auth/callback`	Supabase OAuth callback handler

AI Personas

Built-in personas

Slug	Name	Style	Post types	Media
`lecture-bestie`	Lecture Bestie	Casual, friendly plain English	concept, recap	text, audio
`exam-gremlin`	Exam Gremlin	Mischievous, trap-focused	trap	text
`problem-grinder`	Problem Grinder	Methodical, step-by-step	example	text
`doodle-prof`	Doodle Prof	3-panel comic strip (JSON output)	concept	image
`meme-lord`	Meme Lord	Classic meme templates (JSON output)	concept	image
`study-bard`	Study Bard	Song lyrics with hook/verse/chorus (JSON output)	recap	audio

doodle-prof, meme-lord, and study-bard return structured JSON from the LLM. The sub-workflows parse this into rendered images or audio before saving to Supabase Storage.

Custom personas

Authenticated users can create their own personas via POST /api/personas. Required fields: name, slug, role_tag, tone, teaching_style, description. Optional: emoji, accent_color, is_public. The system prompt is auto-generated from name + role_tag + tone + teaching_style + description. Custom personas can be used for post generation (by including the slug in enabled_personas) and appear in DMs and quiz chat like built-in ones. RLS allows: owner full CRUD, public custom personas visible to all authenticated users, private personas visible only to their owner.

Post Types and Sources

`post_type`	Description
`concept`	Core concept explanation
`example`	Worked example or practice problem
`trap`	Common exam mistake or trick
`review`	Spaced-repetition review card
`recap`	Summary or overview
`pulse`	Persona community update (from Workflow 8)

`source`	Origin
`n8n`	Generated from user-uploaded material (Workflow 1)
`creao`	Ingested from Creao via Google Sheets (Workflow 4)
`reactive`	Remedial post triggered by quiz failures (Workflow 7)
`pulse`	Community update post from Workflow 8

Learner Memory & Companion Layer

Three services work together to provide persistent, personalized learning:

HydraDB stores structured learner memories — topics studied, misconceptions identified, quiz performance, and engagement patterns. Memories are written on quiz submissions, Telegram inbound replies, and material interactions; recalled as context for quiz hints, Telegram responses, and Dify agent runs.

Dify teaching workflow (n8n/dify-teaching-workflow.yml) is a configurable agent graph built in the Dify UI. It receives trigger events (material_uploaded, feed_opened) and decides what to send back — emphasis signals and priority personas (stored in materials.teaching_plan) for content generation, or nudge messages for Telegram. Uses GMI Cloud for inference.

Telegram Bot handles ambient study companion delivery. Students link their Telegram account via a deep-link from their profile. Scrollabus sends proactive nudges; inbound messages are routed to /api/telegram/webhook where intent classification dispatches to the appropriate handler (quiz, recap, simpler explanation, or general GMI chat with memory context).

Environment Variables

Variable	Required	Used by	Notes
`NEXT_PUBLIC_SUPABASE_URL`	Yes	Next.js	Public Supabase project URL
`NEXT_PUBLIC_SUPABASE_ANON_KEY`	Yes	Next.js	Public anon key for client-side auth
`SUPABASE_SERVICE_KEY`	Yes	Next.js API routes	Service-role key; never expose to browser
`GEMINI_API_KEY`	Yes	Next.js	PDF vision parse + quiz generation + quiz chat + DM chat
`ELEVENLABS_API_KEY`	Yes	Next.js `/api/tts`	In-app text-to-speech
`N8N_WEBHOOK_MATERIAL_TO_POST`	Yes	Next.js	Webhook URL for Workflow 1
`N8N_WEBHOOK_COMMENT_REPLY`	Yes	Next.js	Webhook URL for Workflow 2
`YOUTUBE_API_KEY`	No	Next.js + n8n	YouTube Data API v3; needed for `/api/explore/search` (Next.js) and Workflow 5 (n8n); set in both
`HYDRADB_API_KEY`	No	Next.js	Learner memory layer; omit to disable
`HYDRADB_TENANT_ID`	No	Next.js	Defaults to `scrollabus`
`DIFY_API_URL`	No	Next.js	Dify teaching agent base URL (default: `https://api.dify.ai/v1`)
`DIFY_API_KEY`	No	Next.js	Dify workflow API key
`GMI_CLOUD_API_KEY`	No	Next.js	DeepSeek/Llama via GMI for Telegram responses; models: `DeepSeek-V3-0324` (fast), `DeepSeek-R1-0528` (reasoning)
`TELEGRAM_BOT_TOKEN`	No	Next.js	Telegram Bot study companion (from @BotFather)
`FEATHERLESS_API_KEY`	Yes	n8n only	Set as n8n variable, not in `.env.local`
`SUPABASE_URL`	Yes	n8n only	Set as n8n variable (same value as public URL)
`SUPABASE_SERVICE_KEY`	Yes	n8n only	Set as n8n variable
`KLING_API_KEY`	No	n8n only	Kling AI video generation (Workflow 1e); set inside n8n
`GOOGLE_SHEET_ID`	No	n8n Workflows 3 & 4	Sheet ID for Creao bridge

Troubleshooting

Feed is empty after uploading material

Check that Workflow 1 is active in n8n and the webhook URL in .env.local matches exactly.
Open n8n → Executions to see if the workflow ran and inspect any errors.
Posts appear asynchronously — wait a few seconds and refresh.

PDF parse returns empty or truncated text

Gemini vision requires GEMINI_API_KEY. Without it, the fallback is pdf-parse which cannot read scanned or image-only PDFs.
PDFs larger than 10 MB are rejected.

Comments get no AI reply

Verify N8N_WEBHOOK_COMMENT_REPLY is set and Workflow 2 is active.
Check that FEATHERLESS_API_KEY is set as an n8n variable (not in .env.local).
Replies are fire-and-forget; errors are logged server-side but do not surface to the user.

n8n workflow IDs are wrong in Workflow 1

After importing all workflows, open Workflow 1, click each Execute Workflow node, and replace the placeholder workflowId values with the numeric IDs shown in your n8n workflow list.

PGRST200 "relationship not found" errors in feed

This is a PostgREST schema cache issue. Manually reload the Supabase schema cache under Settings → API → Reload schema.

Supabase RLS blocks posts from n8n

n8n uses the service-role key, which bypasses RLS. If you see 403s from n8n, confirm the Authorization: Bearer <service-key> header is set in each HTTP node.

Telegram bot receives no inbound messages

Verify TELEGRAM_BOT_TOKEN is set and the webhook is registered (call POST /api/telegram/setup-webhook once after deployment).
Confirm the student has linked their Telegram (telegram_enabled = true) and their telegram_chat_id is set in the profile record.

Quizzes not generating

GEMINI_API_KEY is required for quiz generation (lib/quizzes.ts uses Gemini 2.5 Flash directly).
Ensure the material raw_text is not empty; quiz generation reads from the material text (truncated to 4000 characters for the LLM prompt).

Reactive or pulse posts not appearing

Verify Workflows 7 and 8 are active in n8n.
These workflows read from Supabase via the service-role key — confirm SUPABASE_SERVICE_KEY and FEATHERLESS_API_KEY are set as n8n variables.

Project Structure

scrollabus/
├── app/
│   ├── (app)/              # Authenticated routes (feed, explore, community, saved, upload, profile)
│   ├── (auth)/             # Login
│   ├── (onboarding)/       # Onboarding flow
│   ├── api/                # Next.js API routes
│   │   ├── comments/
│   │   ├── community/
│   │   ├── explore/
│   │   │   └── search/     # Live YouTube search
│   │   ├── feed/
│   │   ├── impressions/
│   │   ├── likes/
│   │   ├── materials/
│   │   │   └── parse-pdf/
│   │   ├── personas/       # GET list, POST create custom persona
│   │   │   └── [slug]/     # GET, PATCH, DELETE; /dm for DM chat
│   │   ├── telegram/       # connect / webhook / cron / setup-webhook
│   │   ├── posts/[id]/
│   │   ├── profile/
│   │   ├── quiz-chat/
│   │   ├── quiz-responses/
│   │   ├── quizzes/
│   │   │   └── generate/
│   │   ├── save-categories/
│   │   ├── saved-posts/
│   │   ├── saves/
│   │   └── tts/
│   ├── globals.css
│   ├── layout.tsx
│   └── manifest.ts         # PWA manifest
├── components/             # React UI components (feed, explore, quiz, social, onboarding)
├── lib/
│   ├── constants.ts        # Persona config, feed constants, post/media/quiz type display config
│   ├── dify.ts             # Dify teaching agent helpers
│   ├── gmi.ts              # GMI Cloud (OpenAI-compatible) client
│   ├── hydra.ts            # HydraDB learner memory helpers
│   ├── n8n.ts              # Webhook trigger helpers
│   ├── personas.ts         # Static built-in persona definitions (mirrored in DB via seed SQL)
│   ├── telegram.ts         # Telegram Bot API client
│   ├── quizzes.ts          # Quiz generation logic (Gemini 2.5 Flash, persona-voiced)
│   ├── types.ts            # TypeScript interfaces and union types
│   └── supabase/           # Supabase client factories (client, server, service, middleware)
├── n8n/                    # n8n workflow JSON files + Dify workflow YAML exports
│   ├── workflow-1-material-to-posts.json
│   ├── workflow-1{a-e}-sub-*.json  # Sub-workflows: text, image, audio-tts, audio-bard, video
│   ├── workflow-2-comment-reply.json
│   ├── workflow-3-engagement-sync.json
│   ├── workflow-4-creao-ingest.json
│   ├── workflow-5-youtube-fetch.json
│   ├── workflow-6-tiktok-fetch.json
│   ├── workflow-7-reactive-content.json
│   ├── workflow-8-persona-pulse.json
│   ├── dify-teaching-workflow.yml
│   └── dify-video-workflow.yml
├── supabase/
│   ├── schema.sql              # Base schema (run first)
│   ├── seed.sql                # Persona seed data
│   ├── migrations/             # Incremental schema additions
│   ├── engagement-rpc.sql      # get_trending_posts RPC
│   ├── community-rpc.sql       # get_similar_users RPC
│   ├── external-content-rpc.sql
│   └── persona-affinity-rpc.sql  # get_persona_affinity RPC (adaptive feed)
├── types/                  # TypeScript declaration files (e.g. pdf-parse.d.ts)
├── middleware.ts            # Supabase session refresh on every request
├── .env.example
└── README.md

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Repository files navigation

Scrollabus

Features

Tech Stack

Architecture

Data Model

Quickstart

1. Prerequisites

2. Clone and install

3. Configure environment

4. Set up Supabase

5. Import n8n workflows

6. Run

How It Works

Material Upload Flow

Comment Reply Flow

Quiz Generation

Feed Algorithm

Scheduled n8n Workflows

API Routes

AI Personas

Built-in personas

Custom personas

Post Types and Sources

Learner Memory & Companion Layer

Environment Variables

Troubleshooting

Project Structure

About

Uh oh!

Releases

Packages

Uh oh!

Contributors

Uh oh!

Languages

Name		Name	Last commit message	Last commit date
Latest commit History 3 Commits
app		app
components		components
docs		docs
lib		lib
n8n		n8n
public		public
supabase		supabase
types		types
.env.example		.env.example
.gitignore		.gitignore
README.md		README.md
Scrollabus_Logo.png		Scrollabus_Logo.png
middleware.ts		middleware.ts
next.config.ts		next.config.ts
package-lock.json		package-lock.json
package.json		package.json
postcss.config.js		postcss.config.js
tailwind.config.ts		tailwind.config.ts
tsconfig.json		tsconfig.json

Folders and files

Latest commit

History

Repository files navigation

Scrollabus

Features

Tech Stack

Architecture

Data Model

Quickstart

1. Prerequisites

2. Clone and install

3. Configure environment

4. Set up Supabase

5. Import n8n workflows

6. Run

How It Works

Material Upload Flow

Comment Reply Flow

Quiz Generation

Feed Algorithm

Scheduled n8n Workflows

API Routes

AI Personas

Built-in personas

Custom personas

Post Types and Sources

Learner Memory & Companion Layer

Environment Variables

Troubleshooting

Project Structure

About

Resources

Uh oh!

Stars

Watchers

Forks

Releases

Packages 0

Uh oh!

Contributors

Uh oh!

Languages

Packages