Skip to content

tech_stack.md

Latest commit

 

History

History
24 lines (18 loc) · 1.02 KB

tech_stack.md

File metadata and controls

24 lines (18 loc) · 1.02 KB

Hidayah AI Technical Specification

Core Frameworks

  • Frontend: Streamlit (Python)
  • Styling: Tailwind CSS injected via st.markdown(unsafe_allow_html=True)
  • LLM: Google Gemini 1.5 Flash (via google-generativeai)

Key Skills & Logic

  1. Agent Router Logic: - Function: classify_intent(query)

    • Categories: [VERSE_LOOKUP, SCHOLARLY_RESEARCH, PDF_ANALYSIS]

  2. Web Search Tool: - API: Tavily (use tavily-python)

    • Purpose: Fetching historical context and modern scholarly fatwas.

  3. Audio Engine:

    • API: AlQuran.cloud (Endpoint: https://api.alquran.cloud/v1/juz/{juz}/ar.alafasy)
    • Translation Sync: Fetch ur.jalandhry for Urdu and en.asad for English.

  4. RAG (Document Analysis):

    • Libraries: PyPDF2 for parsing, FAISS for vector storage.
    • Flow: Extract PDF text -> Create chunks -> Embed using Gemini -> Query when intent is PDF_ANALYSIS.

  5. State Management:

    • Use st.session_state to track current Juz, playback time, and chat history.