Update app.py

app.py

@@ -165,71 +165,71 @@ def get_file(source_documents):
     return references, files_in_order
 
 
-def build_chain(vectordb: Chroma):
-    system_instruction = (
-        "You are an expert **Electrical Engineer AI Assistant**, specialized in power systems "
-        "and substation design (AIS/GIS up to 765kV), providing insights strictly from the provided context.\n\n"
-        "**Formatting Guidelines:**\n"
-        "1. Organize using **bullet points or numbered lists** where appropriate.\n"
-        "2. **Bold** key technical terms, parameters, and essential facts.\n"
-        "3. Use **technical language** consistent with IEC/IEEE/POWERGRID standards.\n"
-        "4. For multi-step explanations, use **sub-headings** (e.g., `## Sub-section`).\n"
-        "5. **Always include clause references (e.g., Clause XX.XX) for every piece of information.**\n"
-        "6. **CRITICAL: If context contains a table, reproduce it EXACTLY — preserve all rows, "
-        "columns, headers, and alignment. Never paraphrase table data.**\n\n"
-        "**Context Prioritization:**\n"
-        "1. Prioritize documents directly related to the queried equipment type.\n"
-        "2. 'Specific Requirements' clauses **supersede** all other documents — reflect modified clauses first.\n"
-        "3. If context is insufficient: 'The available documents do not contain information regarding [detail].'\n"
-        "4. **Do not invent information** outside the provided context."
-    )
-
-    prompt = ChatPromptTemplate.from_messages([
-        SystemMessagePromptTemplate.from_template(system_instruction),
-        MessagesPlaceholder(variable_name="chat_history"),
-        HumanMessagePromptTemplate.from_template(
-            "Context:\n{context}\n\nQuestion:\n{question}"
-        ),
-    ])
-
-    # ── Groq LLM ───────────────────────────────────────────────────────────────
-    llm = ChatGroq(
-        model=GROQ_MODEL,
-        temperature=0.1,
-        max_tokens=2048,
-        api_key=GROQ_API_KEY,
-    )
-
-    # ── Retriever ──────────────────────────────────────────────────────────────
-    retriever = vectordb.as_retriever(
-        search_type="mmr",
-        search_kwargs={"k": 3, "lambda_mult": 0.5, "fetch_k": 15},
-    )
-
-    def format_docs(docs):
-        return "\n\n---\n\n".join(doc.page_content for doc in docs)
-
-    rag_core = (
-        RunnablePassthrough.assign(
-            context=lambda x: format_docs(retriever.invoke(x["question"]))
-        )
-        | prompt
-        | llm
-        | StrOutputParser()
-    )
-
-    chain_with_history = RunnableWithMessageHistory(
-        rag_core,
-        get_session_history,
-        input_messages_key="question",
-        history_messages_key="chat_history",
-    )
-
-    return chain_with_history, retriever
-
-
-# ── Build once at startup (not per Gradio call) ───────────────────────────────
-chain, retriever = build_chain(vectordb)   # vectordb initialised elsewhere
+# def build_chain(vectordb: Chroma):
+#     system_instruction = (
+#         "You are an expert **Electrical Engineer AI Assistant**, specialized in power systems "
+#         "and substation design (AIS/GIS up to 765kV), providing insights strictly from the provided context.\n\n"
+#         "**Formatting Guidelines:**\n"
+#         "1. Organize using **bullet points or numbered lists** where appropriate.\n"
+#         "2. **Bold** key technical terms, parameters, and essential facts.\n"
+#         "3. Use **technical language** consistent with IEC/IEEE/POWERGRID standards.\n"
+#         "4. For multi-step explanations, use **sub-headings** (e.g., `## Sub-section`).\n"
+#         "5. **Always include clause references (e.g., Clause XX.XX) for every piece of information.**\n"
+#         "6. **CRITICAL: If context contains a table, reproduce it EXACTLY — preserve all rows, "
+#         "columns, headers, and alignment. Never paraphrase table data.**\n\n"
+#         "**Context Prioritization:**\n"
+#         "1. Prioritize documents directly related to the queried equipment type.\n"
+#         "2. 'Specific Requirements' clauses **supersede** all other documents — reflect modified clauses first.\n"
+#         "3. If context is insufficient: 'The available documents do not contain information regarding [detail].'\n"
+#         "4. **Do not invent information** outside the provided context."
+#     )
+
+#     prompt = ChatPromptTemplate.from_messages([
+#         SystemMessagePromptTemplate.from_template(system_instruction),
+#         MessagesPlaceholder(variable_name="chat_history"),
+#         HumanMessagePromptTemplate.from_template(
+#             "Context:\n{context}\n\nQuestion:\n{question}"
+#         ),
+#     ])
+
+#     # ── Groq LLM ───────────────────────────────────────────────────────────────
+#     llm = ChatGroq(
+#         model=GROQ_MODEL,
+#         temperature=0.1,
+#         max_tokens=2048,
+#         api_key=GROQ_API_KEY,
+#     )
+
+#     # ── Retriever ──────────────────────────────────────────────────────────────
+#     retriever = vectordb.as_retriever(
+#         search_type="mmr",
+#         search_kwargs={"k": 3, "lambda_mult": 0.5, "fetch_k": 15},
+#     )
+
+#     def format_docs(docs):
+#         return "\n\n---\n\n".join(doc.page_content for doc in docs)
+
+#     rag_core = (
+#         RunnablePassthrough.assign(
+#             context=lambda x: format_docs(retriever.invoke(x["question"]))
+#         )
+#         | prompt
+#         | llm
+#         | StrOutputParser()
+#     )
+
+#     chain_with_history = RunnableWithMessageHistory(
+#         rag_core,
+#         get_session_history,
+#         input_messages_key="question",
+#         history_messages_key="chat_history",
+#     )
+
+#     return chain_with_history, retriever
+
+
+# # ── Build once at startup (not per Gradio call) ───────────────────────────────
+# chain, retriever = build_chain(vectordb)   # vectordb initialised elsewhere
 
 
 # Query Re-write
@@ -242,17 +242,17 @@ def rewrite_query(question: str, llm) -> str:
     rewrite_prompt = PromptTemplate.from_template("""
 You are an expert query rewriter for a POWERGRID technical document retrieval system.
 The document corpus contains:
-- Model Technical Specifications for GIS/AIS substations (220kV / 400kV / 765kV)
+- Model Technical Specifications for various equipments used in GIS/AIS substations (132kV /220kV / 400kV / 765kV)
 - IEC and IEEE standards referenced in POWERGRID specs
-- Equipment-specific specs: Circuit Breakers, Isolators, Surge Arresters, CTs, VTs, 
-  Power Transformers, Reactors, Protection Relays, Control & Relay Panels
-- Specific Requirements documents (which supersede other docs)
+- Equipment-specific specs: Circuit Breakers, Isolators, Surge Arresters, CTs, VTs, Gas Insulated Switchgears,
+  Power Transformers, Reactors, Protection Relays, Control & Relay Panels, Visual Monitoring Systems (VMS), Switchyard Erection 
+- Specific Requirements Document (which supersede other docs)
 
 Your task:
 1. Expand abbreviations (e.g., CB → Circuit Breaker, SA → Surge Arrester, CT → Current Transformer)
 2. Add relevant technical keywords likely present in the documents
 3. Include clause/section indicators if the query implies a specific requirement
-4. If the query is vague, make it specific to power system substation context
+4. If the query is vague, make it specific to power system Substation context
 5. Preserve the original intent — do NOT change what is being asked
 6. Output ONLY the rewritten query, nothing else