Skip to content

instructions.md

Latest commit

 

History

History
151 lines (106 loc) · 3.45 KB

instructions.md

File metadata and controls

151 lines (106 loc) · 3.45 KB
Raw

Dyut Website Research Agent (PRAT Framework)

Product + Technical Specification (Next.js Frontend + Python Backend)

Goal: A local-only web dashboard for YouTubers that generates actionable trend and content opportunity research from any URL (optimized for YouTube + Reddit) and outputs a single Markdown report per request.

Architecture Overview

  • Frontend: Next.js (Dashboard UI)
  • Backend: Python (FastAPI)
  • Execution: Synchronous request/response
  • Storage: Local JSON file (append-only)
  • No authentication
  • No official APIs
  • No citations in output

1. PRAT Framework

P --- Perceive

  • Parse user prompt
  • Extract keywords
  • Classify intent (trend discovery, influencer ranking, content ideation)
  • Expand semantic keyword set
  • Build research plan

R --- Reason

  • Determine source strategy (YouTube, Reddit, generic web)
  • Expand search space
  • Apply time filters
  • Score candidates based on:
    • Engagement proxy
    • Recency
    • Keyword match

A --- Act

  • Scrape pages
  • Normalize into unified ContentItem schema
  • Rank items
  • Generate structured Markdown report

T --- Track

  • Append each run to local JSON store
  • Store:
    • Inputs
    • Extracted keywords
    • Ranked results
    • Generated Markdown

2. Backend (Python - FastAPI)

Folder Structure

backend/ │ ├── app/ │ ├── main.py │ ├── routes/ │ │ └── research.py │ ├── core/ │ │ ├── pipeline.py │ │ ├── ranking.py │ │ ├── markdown.py │ │ ├── storage.py │ │ └── errors.py │ ├── sources/ │ │ ├── base.py │ │ ├── youtube.py │ │ ├── reddit.py │ │ └── generic.py │ └── data/ └── research_history.json

API Endpoint

POST /api/research

Request: - target_urls: list[str] - prompt: str - time_window: optional (24h, 7d, 14d, 30d) - category: optional - num_results: 1--20 - include_debug: bool

Response: - report_markdown: str - results: structured ranked list - stored_record_id: str

3. Data Model

ContentItem

  • id
  • source (youtube | reddit | generic)
  • url
  • title
  • author
  • published_at
  • extracted_text
  • engagement metrics
  • raw metadata

ResearchRunRecord

  • id
  • created_at
  • inputs
  • plan
  • selected_results
  • report_markdown

4. Frontend (Next.js)

Routes

  • /dashboard
  • /history
  • /history/[id]

Components

  • ResearchForm
  • RunStatus
  • MarkdownViewer
  • HistoryTable

UX Requirements

  • Inline validation
  • Presets
  • Copy Markdown
  • Download .md
  • Clear error handling

5. Local Deployment

Backend: uvicorn app.main:app --reload --port 8000

Frontend: npm run dev

6. Acceptance Criteria

  • Synchronous Markdown research generation
  • Structured output
  • Persistent JSON history
  • Fully local execution