Update src/ingestion.py

src/ingestion.py

@@ -3,7 +3,6 @@ import fitz  # PyMuPDF
 import unicodedata
 import os
 import json
-from gen_ai_hub.proxy.langchain.openai import ChatOpenAI  # ✅ use SAP GenAI Hub LLM
 
 # ==========================================================
 # 1️⃣ TEXT EXTRACTION (Clean + TOC Detection)
@@ -126,14 +125,15 @@ def extract_table_of_contents(text: str):
 # ==========================================================
 # 3A️⃣ HYBRID TOC FALLBACK (AI-Inferred using SAP GenAI Hub)
 # ==========================================================
-def adaptive_fallback_toc(text: str, model: str = "gpt-4o-mini", max_chars: int = 7000):
+import requests
+
+def adaptive_fallback_toc(text: str, max_chars: int = 7000):
     """
-    Uses SAP GenAI Hub LLM to infer a Table of Contents from document text.
-    Reads client_id/secret/deployment_name from JSON credentials file.
+    Uses SAP GenAI Hub REST API directly (client credentials token flow)
+    to infer a Table of Contents from document text.
     """
     snippet = text[:max_chars]
 
-    # ✅ Load GenAI credentials JSON
     creds_path = os.path.join(os.path.dirname(__file__), "GEN AI HUB PROXY.json")
     if not os.path.exists(creds_path):
         print("⚠️ No SAP GenAI credentials file found — skipping AI fallback.")
@@ -142,11 +142,31 @@ def adaptive_fallback_toc(text: str, model: str = "gpt-4o-mini", max_chars: int
     with open(creds_path) as f:
         creds = json.load(f)
 
-    deployment_name = creds.get("deployment_name", model)
-    print(f"🔑 Using GenAI deployment: {deployment_name}")
+    client_id = creds.get("client_id")
+    client_secret = creds.get("client_secret")
+    token_url = creds.get("token_url")
+    base_url = creds.get("base_url", "").rstrip("/")
+    deployment_name = creds.get("deployment_name", "gpt-4o-mini")
+
+    if not all([client_id, client_secret, token_url, base_url]):
+        print("⚠️ Missing fields in GEN AI HUB PROXY.json — skipping AI fallback.")
+        return []
 
     try:
-        llm = ChatOpenAI(model=deployment_name, temperature=0)
+        # 1️⃣ Get token
+        token_resp = requests.post(
+            token_url,
+            data={"grant_type": "client_credentials"},
+            auth=(client_id, client_secret),
+        )
+        token_resp.raise_for_status()
+        token = token_resp.json().get("access_token")
+
+        # 2️⃣ Call SAP GenAI deployment
+        headers = {
+            "Authorization": f"Bearer {token}",
+            "Content-Type": "application/json",
+        }
 
         prompt = f"""
         You are a document structure analyzer.
@@ -156,13 +176,32 @@ def adaptive_fallback_toc(text: str, model: str = "gpt-4o-mini", max_chars: int
         TEXT SAMPLE:
         {snippet}
         """
-        response = llm.invoke(prompt)
+
+        body = {
+            "model": deployment_name,
+            "input": prompt
+        }
+
+        endpoint = f"{base_url}/v2/inference/deployments/{deployment_name}/responses"
+        response = requests.post(endpoint, headers=headers, json=body)
+        response.raise_for_status()
+        data = response.json()
+
+        # Extract text safely from different SAP formats
+        content = ""
+        if isinstance(data, dict):
+            if "choices" in data and len(data["choices"]) > 0:
+                content = data["choices"][0].get("message", {}).get("content", "")
+            elif "output" in data:
+                content = data["output"][0]["content"][0]["text"]
+
         lines = [
             re.sub(r"^[0-9.\-•\\s]+", "", l.strip())
-            for l in response.content.splitlines()
+            for l in content.splitlines()
             if l.strip()
         ]
         toc_ai = [(str(i + 1), l) for i, l in enumerate(lines) if len(l) > 3]
+        print(f"✨ AI-inferred TOC generated with {len(toc_ai)} entries.")
         return toc_ai
 
     except Exception as e: