ASISTEN PROKTOR ASESMEN NASIONAL

🖼 Arsitektur Chat Bot

Framework Code Igniter 4

Frontend (User) • Browser (Bootstrap 5 UI)
↕
Backend (CI4 - PHP)
  • Controller:
    o Upload PDF → Parser → Chunking → Store ke DB
    o Chat Query → buat embedding pertanyaan → similarity search
  • Services:
    o PDF Parser & Chunker
    o Embedding API client
    o LLM API client
    o Enrichment

Database (MySQL/MariaDB)
  • Tabel:
    o documents (info PDF)
    o chunks (potongan teks PDF)
    o embeddings (vektor teks JSON/longtext)
    o chat_history (multi-turn chat)

Layanan Pihak Ketiga
  • HuggingFace Inference API → generate embeddings
  • Groq API → generate jawaban
  • Wikipedia REST API → enrichment tambahan

───────────────────────────────
🔄 Alur Data
1. Upload PDF → CI4 parser → potong jadi chunk → embedding (HuggingFace) → simpan ke DB.
2. User bertanya → buat embedding pertanyaan → similarity search (cosine similarity manual).
3. Ambil top-N context chunks → compose prompt → kirim ke Groq API.
4. Jika skor similarity rendah → panggil Wikipedia API → gabungkan ke prompt.
5. Jawaban ditampilkan ke user (Bootstrap).