Upload folder using huggingface_hub

modules/llm/summarizer/helpers/db_helper.py

@@ -0,0 +1,57 @@
+import uuid
+from config import SanatanConfig
+from db import SanatanDatabase
+from modules.llm.summarizer.models import ScriptureRequest
+import logging
+
+logging.basicConfig()
+logger = logging.getLogger(__name__)
+logger.setLevel(logging.INFO)
+
+
+async def get_scripture_from_db(req: ScriptureRequest):
+    """
+    Return a scripture unit (page or verse, based on config),
+    including all metadata fields separately.
+    Used for page view to fetch by global index.
+    """
+    # ensure we have a valid request id (deviceId)
+    request_id = req.request_id or f"auto-{uuid.uuid4()}"
+    logger.info(
+        f"get_scripture: received requestId={request_id}, scripture={req.scripture_name}, unit_index={req.unit_index}"
+    )
+
+    # find config entry for the scripture
+    config = next(
+        (s for s in SanatanConfig().scriptures if s["name"] == req.scripture_name), None
+    )
+    if not config:
+        return {
+            "error": f"Scripture '{req.scripture_name}' not found",
+            "requestId": request_id,
+        }
+
+    # fetch the raw document from DB
+    raw_doc = SanatanDatabase().fetch_document_by_index(
+        collection_name=config["collection_name"],
+        index=req.unit_index,
+    )
+
+    if not raw_doc or isinstance(raw_doc, str) or "error" in raw_doc:
+        return {
+            "error": f"No data available for unit {req.unit_index}",
+            "requestId": request_id,
+        }
+
+    # canonicalize it
+    canonical_doc = SanatanConfig().canonicalize_document(
+        scripture_name=req.scripture_name,
+        document_text=raw_doc.get("document", ""),
+        metadata_doc=raw_doc,
+    )
+
+    # add unit index & total units (so Flutter can paginate)
+    canonical_doc["total"] = SanatanDatabase().count(config["collection_name"])
+    canonical_doc["requestId"] = request_id
+
+    return canonical_doc

modules/llm/summarizer/helpers/llm_helper.py

@@ -0,0 +1,53 @@
+from openai import OpenAI
+
+from modules.llm.summarizer.models import ScriptureVerseSummary
+
+
+client = OpenAI()
+
+
+def summarize_scripture_verse(
+    lyrics: str, translation: str, word_by_word_meaning: str, target_language: str
+) -> ScriptureVerseSummary:
+    """
+    Generates a simple and detailed meaning of the given lyrics
+    in the specified target language using structured output.
+    """
+    print("Summarizing ...")
+    system_prompt = (
+        "You are a precise multilingual assistant that summarizes meanings of Sanatan scripture verses.\n"
+        "Your job is to derive the verse’s simple and detailed meaning *strictly* from the provided inputs.\n\n"
+        "=== MANDATORY RULES ===\n"
+        "1. Use ONLY the information from: lyrics, translation, and word-by-word meaning.\n"
+        "2. Do NOT invent, guess, or infer anything not explicitly given.\n"
+        "3. Do NOT include any English words or transliterations unless they already appear in the input.\n"
+        "4. Every part of your output must be written completely in the target language: {target_language}.\n"
+        "   - If the target language lacks a direct word, explain it *in that language* (do not leave English placeholders).\n"
+        "   - Preserve transliterated proper nouns (e.g., ‘nanjIyar’, ‘rAmAnuja’) exactly as written — case-sensitive.\n"
+        "5. NEVER mix scripts. Do not output Latin letters unless they occur verbatim in the input.\n"
+        "6. If unsure of a word’s meaning, omit it or express uncertainty *in the target language*.\n"
+        "7. Maintain a respectful, neutral tone. No opinions, commentary, or theological judgment.\n"
+        "8. Output must follow the MeaningResponse schema exactly: only 'simple_meaning', 'detailed_meaning', and 'language'.\n"
+        "9. Do NOT repeat the input text or provide explanations in any other language.\n"
+        "10. Ensure all formatting is plain text — no quotes, brackets, or markdown around meanings.\n"
+    )
+
+    user_prompt = (
+        f"Lyrics:\n{lyrics}\n\n"
+        f"Translation:\n{translation}\n\n"
+        f"Word-by-word meaning:\n{word_by_word_meaning}\n\n"
+        f"Now generate the verse’s meaning entirely in **{target_language}**.\n"
+        "Provide both a simple summary and a detailed explanation in that language.\n"
+        "Do not include any English words unless they appear exactly as in the inputs."
+    )
+
+    response = client.chat.completions.parse(
+        model="gpt-4o-mini",
+        messages=[
+            {"role": "system", "content": system_prompt},
+            {"role": "user", "content": user_prompt},
+        ],
+        response_format=ScriptureVerseSummary,
+    )
+
+    return response.choices[0].message.parsed

modules/llm/summarizer/models.py

@@ -0,0 +1,16 @@
+# --- Define output schema ---
+from typing import Optional
+from pydantic import BaseModel, Field
+
+class ScriptureRequest(BaseModel):
+    scripture_name: str
+    unit_index: int
+    request_id: str | None = Field(
+        default=None, alias="request_id"
+    )  # optional, backward compatible
+    target_language: Optional[str] = Field(default="English")
+
+class ScriptureVerseSummary(BaseModel):
+    target_language: str
+    simple_meaning: str
+    detailed_meaning: str

modules/llm/summarizer/service.py

@@ -0,0 +1,75 @@
+from modules.llm.summarizer.helpers.db_helper import get_scripture_from_db
+from modules.llm.summarizer.helpers.llm_helper import summarize_scripture_verse
+from modules.llm.summarizer.models import ScriptureRequest, ScriptureVerseSummary
+from google.cloud import firestore
+import os
+import json
+from google.oauth2 import service_account
+
+# Load Firebase credentials from environment variable
+service_account_info = os.getenv("FIREBASE_SERVICE_ACCOUNT_JSON")
+
+if not service_account_info:
+    raise ValueError("FIREBASE_SERVICE_ACCOUNT_JSON not found in environment variables.")
+
+# Parse the JSON (it’s stored as a string in env)
+credentials_dict = json.loads(service_account_info)
+credentials = service_account.Credentials.from_service_account_info(credentials_dict)
+
+# Initialize Firestore client with these credentials
+db = firestore.AsyncClient(credentials=credentials, project=credentials.project_id)
+# or db = firestore.Client(credentials=credentials, project=credentials.project_id)
+
+
+async def svc_summarize_scripture_verse(req: ScriptureRequest) -> ScriptureVerseSummary:
+    """
+    Summarizes a scripture verse, with Firestore caching per verse and language.
+    Firestore structure:
+      /scripture_summaries/{scripture_name_global_index}/meanings/{target_language}
+    """
+
+    # Step 1: Get scripture data
+    scripture_data = await get_scripture_from_db(req)
+
+    scripture_name = scripture_data.get("scripture_name", "UnknownScripture")
+    global_index = scripture_data.get("_global_index", -1)
+    target_language = req.target_language.lower()
+
+    # Compose the document path
+    doc_id = f"{scripture_name}_{global_index}".replace(" ", "_")
+    base_ref = db.collection("scripture_summaries").document(doc_id)
+    meaning_ref = base_ref.collection("meanings").document(target_language)
+
+    # Step 2: Check for cached version
+    cached_doc = await meaning_ref.get()
+    if cached_doc.exists:
+        print(f"✅ Using cached summary for {scripture_name} {global_index} ({target_language})")
+        return ScriptureVerseSummary(**cached_doc.to_dict())
+
+    print(f"🧠 Generating new summary for {scripture_name} {global_index} ({target_language})")
+
+    # Step 3: Generate new summary with LLM
+    summary = summarize_scripture_verse(
+        scripture_data.get("text","--no lyrics available--"),
+        scripture_data.get("translation","--no translation available--"),
+        scripture_data.get("word_by_word_native","--no word-by-word meaning available--"),
+        target_language,
+    )
+
+    summary_data = summary.model_dump()
+
+    # Step 4: Ensure parent document exists
+    await base_ref.set(
+        {
+            "scripture_name": scripture_name,
+            "global_index": global_index,
+        },
+        merge=True,
+    )
+
+    # Step 5: Store meaning under nested language doc
+    await meaning_ref.set(summary_data)
+
+    print(f"✅ Stored summary for {scripture_name} {global_index} in Firestore")
+
+    return summary

server.py

@@ -18,6 +18,9 @@ from modules.config.categories import get_scripture_categories
 from modules.dropbox.discources import get_discourse_by_id, get_discourse_summaries
 from modules.firebase.messaging import FcmRequest, fcm_service
 from modules.languages.get_v2 import handle_fetch_languages_v2
+from modules.llm.summarizer.helpers.db_helper import get_scripture_from_db
+from modules.llm.summarizer.models import ScriptureRequest
+from modules.llm.summarizer.service import svc_summarize_scripture_verse
 from modules.quiz.answer_validator import validate_answer
 from modules.quiz.models import Question
 from modules.quiz.quiz_helper import generate_question
@@ -218,62 +221,10 @@ async def handle_get_scriptures():
     return return_values
 
 
-class ScriptureRequest(BaseModel):
-    scripture_name: str
-    unit_index: int
-    request_id: str | None = Field(
-        default=None, alias="request_id"
-    )  # optional, backward compatible
-
-
 @router.post("/scripture")
 async def get_scripture(req: ScriptureRequest):
-    """
-    Return a scripture unit (page or verse, based on config),
-    including all metadata fields separately.
-    Used for page view to fetch by global index.
-    """
-    # ensure we have a valid request id (deviceId)
-    request_id = req.request_id or f"auto-{uuid.uuid4()}"
-    logger.info(
-        f"get_scripture: received requestId={request_id}, scripture={req.scripture_name}, unit_index={req.unit_index}"
-    )
-
-    # find config entry for the scripture
-    config = next(
-        (s for s in SanatanConfig().scriptures if s["name"] == req.scripture_name), None
-    )
-    if not config:
-        return {
-            "error": f"Scripture '{req.scripture_name}' not found",
-            "requestId": request_id,
-        }
-
-    # fetch the raw document from DB
-    raw_doc = SanatanDatabase().fetch_document_by_index(
-        collection_name=config["collection_name"],
-        index=req.unit_index,
-    )
-
-    if not raw_doc or isinstance(raw_doc, str) or "error" in raw_doc:
-        return {
-            "error": f"No data available for unit {req.unit_index}",
-            "requestId": request_id,
-        }
-
-    # canonicalize it
-    canonical_doc = SanatanConfig().canonicalize_document(
-        scripture_name=req.scripture_name,
-        document_text=raw_doc.get("document", ""),
-        metadata_doc=raw_doc,
-    )
-
-    # add unit index & total units (so Flutter can paginate)
-    canonical_doc["total"] = SanatanDatabase().count(config["collection_name"])
-    canonical_doc["requestId"] = request_id
-
-    return canonical_doc
-
+    response = await get_scripture_from_db(req)
+    return response
 
 @router.get("/scripture_configs")
 async def get_scripture_configs():
@@ -653,4 +604,9 @@ async def send_fcm_endpoint(
     if x_admin_key != ADMIN_KEY:
         raise HTTPException(status_code=403, detail="Unauthorized")
 
-    return await fcm_service.send_fcm(request)    
+    return await fcm_service.send_fcm(request)    
+
+@router.post("/summarize_scripture_verse")
+async def summarize_scripture_verse(req: ScriptureRequest):
+    response = await svc_summarize_scripture_verse(req)
+    return response

static/flutter_bootstrap.js

@@ -38,6 +38,6 @@ _flutter.buildConfig = {"engineRevision":"ddf47dd3ff96dbde6d9c614db0d7f019d7c7a2
 
 _flutter.loader.load({
   serviceWorkerSettings: {
-    serviceWorkerVersion: "4183499183"
+    serviceWorkerVersion: "196864405"
   }
 });

static/flutter_service_worker.js

@@ -17,7 +17,7 @@ const RESOURCES = {"assets/AssetManifest.bin": "1b6e81d215d35b84735e3bac9f9afaad
 "assets/assets/scriptures/default.jpg": "34bb646134c531b02c4b2fdd29119881",
 "assets/assets/thiruman_banner.jpg": "34bb646134c531b02c4b2fdd29119881",
 "assets/FontManifest.json": "d4a8d45844bfcdc2b484bfd5676f4e35",
-"assets/fonts/MaterialIcons-Regular.otf": "61ce5927c2813ade6579d2828168189f",
+"assets/fonts/MaterialIcons-Regular.otf": "9c121c758ac4d0dc351f3018d8d14170",
 "assets/NOTICES": "81bb8caa5fb1d366bb6a83bcbc62b753",
 "assets/packages/cupertino_icons/assets/CupertinoIcons.ttf": "33b7d9392238c04c131b6ce224e13711",
 "assets/packages/flutter_inappwebview/assets/t_rex_runner/t-rex.css": "5a8d0222407e388155d7d1395a75d5b9",
@@ -40,16 +40,16 @@ const RESOURCES = {"assets/AssetManifest.bin": "1b6e81d215d35b84735e3bac9f9afaad
 "canvaskit/skwasm_heavy.wasm": "8034ad26ba2485dab2fd49bdd786837b",
 "favicon.png": "74afc5494e90462fd4f3903ec8271c53",
 "flutter.js": "888483df48293866f9f41d3d9274a779",
-"flutter_bootstrap.js": "0eb96612c0a40fd5f45fecbf05b77346",
+"flutter_bootstrap.js": "6aae013c74d63e80fa365a49ddcb55e9",
 "icons/Icon-192.png": "ac9a721a12bbc803b44f645561ecb1e1",
 "icons/Icon-512.png": "96e752610906ba2a93c65f8abe1645f1",
 "icons/Icon-maskable-192.png": "c457ef57daa1d16f64b27b786ec2ea3c",
 "icons/Icon-maskable-512.png": "301a7604d45b3e739efc881eb04896ea",
 "index.html": "155745e967e3e0ecf6d7333ea445658e",
 "/": "155745e967e3e0ecf6d7333ea445658e",
-"main.dart.js": "da6fc9085ee616f181a6ed95aac529e5",
+"main.dart.js": "9bcfbbc9bd3e4984d982faa4bacd39dd",
 "manifest.json": "9d43d4621f8c3ed75ad4ffe729655ecc",
-"version.json": "a4d3746eceb50edb15ef0d9db3a84e52"};
+"version.json": "397f7fc35ae7b7eb4a183c4d944fd6d5"};
 // The application shell files that are downloaded before a service worker can
 // start.
 const CORE = ["main.dart.js",

static/version.json

@@ -1 +1 @@
-{"app_name":"bhashyam_ai","version":"1.0.91","build_number":"128","package_name":"bhashyam_ai"}
+{"app_name":"bhashyam_ai","version":"1.0.91","build_number":"133","package_name":"bhashyam_ai"}