Updated the file structure

app.py

@@ -61,8 +61,8 @@ def _get_engine():
 
     try:
         print("Initialising FinRAG engine (first request)...")
-        from app.llm.fin_rag_engine import FinRAGEngine
-        from app.data.retrieval.filing_resolver import FilingResolver
+        from llm.fin_rag_engine import FinRAGEngine
+        from data.retrieval.filing_resolver import FilingResolver
         _engine   = FinRAGEngine()
         _resolver = FilingResolver()
         print("Engine ready.")
@@ -107,12 +107,12 @@ def extract_filing_id(choice: str) -> Optional[str]:
 
 def chat(
     message:       str,
-    history:       List[dict],
+    history:       List[Tuple[str, str]],
     mode:          str,
     filing_choice: str,
     top_k:         int,
     request:       gr.Request,
-) -> Tuple[List[dict], str, str]:
+) -> Tuple[List[Tuple[str, str]], str, str]:
     """Returns (updated_history, sources_text, cleared_input)."""
 
     if not message.strip():
@@ -127,16 +127,13 @@ def chat(
             f"You have reached the daily limit of **{MAX_QUERIES_PER_DAY} queries**. "
             f"Resets at {reset_time.strftime('%H:%M UTC')}."
         )
-        return history + [{"role": "user", "content": message},
-                          {"role": "assistant", "content": msg}], "", ""
+        return history + [(message, msg)], "", ""
 
     # Load engine
     try:
         engine, resolver = _get_engine()
     except Exception as e:
-        msg = f"Service unavailable: {e}"
-        return history + [{"role": "user", "content": message},
-                          {"role": "assistant", "content": msg}], "", ""
+        return history + [(message, f"Service unavailable: {e}")], "", ""
 
     filing_id = extract_filing_id(filing_choice)
 
@@ -154,31 +151,24 @@ def chat(
             ref = arg or filing_id
             if not ref:
                 reply = "[X] Specify a company: `/risks NVIDIA` or select a filing from the dropdown."
-                return history + [{"role": "user", "content": message},
-                                  {"role": "assistant", "content": reply}], "", ""
+                return history + [(message, reply)], "", ""
             resp    = engine.summarise_risks(ref)
             sources = ", ".join(resp.source_filings) or "N/A"
-            return (history + [{"role": "user", "content": message},
-                                {"role": "assistant", "content": resp.answer}],
-                    f"Sources: {sources}", "")
+            return history + [(message, resp.answer)], f"Sources: {sources}", ""
 
         elif cmd == "/financials":
             ref = arg or filing_id
             if not ref:
                 reply = "[X] Specify a company: `/financials MSFT` or select a filing."
-                return history + [{"role": "user", "content": message},
-                                  {"role": "assistant", "content": reply}], "", ""
+                return history + [(message, reply)], "", ""
             resp    = engine.extract_financials(ref)
             sources = ", ".join(resp.source_filings) or "N/A"
-            return (history + [{"role": "user", "content": message},
-                                {"role": "assistant", "content": resp.answer}],
-                    f"Sources: {sources}", "")
+            return history + [(message, resp.answer)], f"Sources: {sources}", ""
 
         elif cmd == "/compare":
             if not arg:
                 reply = "[X] Usage: `/compare NVIDIA, Microsoft | AI revenue`"
-                return history + [{"role": "user", "content": message},
-                                  {"role": "assistant", "content": reply}], "", ""
+                return history + [(message, reply)], "", ""
             if "|" in arg:
                 companies_raw, topic = arg.split("|", 1)
             else:
@@ -187,9 +177,7 @@ def chat(
             companies = [c.strip() for c in companies_raw.split(",") if c.strip()]
             resp      = engine.compare_companies(companies, topic.strip())
             sources   = ", ".join(resp.source_filings) or "N/A"
-            return (history + [{"role": "user", "content": message},
-                                {"role": "assistant", "content": resp.answer}],
-                    f"Sources: {sources}", "")
+            return history + [(message, resp.answer)], f"Sources: {sources}", ""
 
         elif cmd == "/help":
             help_text = (
@@ -200,8 +188,7 @@ def chat(
                 "- `/compare <co1>, <co2> | <topic>` — compare companies\n\n"
                 "Or just ask naturally: *What are Google's main revenue segments?*"
             )
-            return (history + [{"role": "user", "content": message},
-                                {"role": "assistant", "content": help_text}], "", "")
+            return history + [(message, help_text)], "", ""
 
     # ── Normal question ──────────────────────────────────────────────── #
     resp    = engine.ask(
@@ -213,8 +200,7 @@ def chat(
     )
     sources = ", ".join(resp.source_filings) if resp.source_filings else "N/A"
     return (
-        history + [{"role": "user",      "content": message},
-                   {"role": "assistant", "content": resp.answer}],
+        history + [(message, resp.answer)],
         f"Sources: {sources}  |  Mode: {resp.retrieval_mode}  |  {_usage_text(request)}",
         "",
     )
@@ -232,8 +218,7 @@ def quick_risks(history, filing, request: gr.Request):
     fid = extract_filing_id(filing)
     if not fid:
         msg = "Please select a filing from the dropdown first."
-        return (history + [{"role": "user", "content": "/risks"},
-                            {"role": "assistant", "content": msg}], "")
+        return history + [("/risks", msg)], ""
     new_h, src, _ = chat(f"/risks {fid}", history, "Local", filing, 10, request)
     return new_h, src
 
@@ -242,8 +227,7 @@ def quick_financials(history, filing, request: gr.Request):
     fid = extract_filing_id(filing)
     if not fid:
         msg = "Please select a filing from the dropdown first."
-        return (history + [{"role": "user", "content": "/financials"},
-                            {"role": "assistant", "content": msg}], "")
+        return history + [("/financials", msg)], ""
     new_h, src, _ = chat(f"/financials {fid}", history, "Local", filing, 10, request)
     return new_h, src
 
@@ -395,7 +379,7 @@ COMMANDS_MD = """
 def build_ui():
     filing_choices = get_filing_choices()
 
-    with gr.Blocks(title="FinSight", css=CSS) as demo:
+    with gr.Blocks(title="FinSight") as demo:
 
         gr.HTML("""
         <div class="fin-header">
@@ -414,7 +398,6 @@ def build_ui():
                     show_label=False,
                     elem_classes=["chatbot-wrap"],
                     render_markdown=True,
-                    type="messages",
                     placeholder=(
                         "<div style='text-align:center;color:#1e2530;"
                         "font-family:DM Mono,monospace;font-size:12px;padding:60px 20px'>"
@@ -524,4 +507,5 @@ if __name__ == "__main__":
         server_name="0.0.0.0",
         server_port=7860,
         show_error=True,
+        css=CSS,
     )

{app/data → data}/retrieval/filing_resolver.py

@@ -1,5 +1,5 @@
 from typing import Optional, List, Dict
-from app.data.retrieval.graph_retriever import GraphRetriever
+from data.retrieval.graph_retriever import GraphRetriever
 
 
 class FilingResolver:

{app/data → data}/retrieval/graph_retriever.py

@@ -1,7 +1,7 @@
 from neo4j import GraphDatabase
 from typing import List, Dict, Any, Optional
 
-from app.config import settings
+from config import settings
 
 
 class GraphRetriever:

{app/data → data}/retrieval/hybridrag_retriever.py

@@ -1,10 +1,10 @@
 from dataclasses import dataclass, field
 from typing import List, Dict, Any, Optional
 
-from app.config import settings
-from app.data.retrieval.reranker import Reranker
-from app.data.retrieval.graph_retriever import GraphRetriever
-from app.data.retrieval.weaviate_retriever import WeaviateRetriever
+from config import settings
+from data.retrieval.reranker import Reranker
+from data.retrieval.graph_retriever import GraphRetriever
+from data.retrieval.weaviate_retriever import WeaviateRetriever
 
 
 @dataclass

{app/data → data}/retrieval/reranker.py

@@ -17,7 +17,7 @@ class Reranker:
 
         if self.use_cross_encoder:
             try:
-                from app.utils.reranker_utils import NimReranker
+                from utils.reranker_utils import NimReranker
                 self._ce_client = NimReranker()
             except Exception:
                 self.use_cross_encoder = False

{app/data → data}/retrieval/weaviate_retriever.py

@@ -3,8 +3,8 @@ from weaviate.classes.init import Auth
 from weaviate.classes.query import Filter, MetadataQuery
 from typing import List, Dict, Any, Optional
 
-from app.config import settings
-from app.utils.embedding_utils import BGEM3Embedder
+from config import settings
+from utils.embedding_utils import BGEM3Embedder
 
 
 class WeaviateRetriever:

{app/llm → llm}/fin_rag_engine.py

@@ -1,9 +1,9 @@
 from dataclasses import dataclass, field
 from typing import List, Dict, Any, Optional, Iterator
 
-from app.llm.groq_client import GroqClient
-from app.data.retrieval.filing_resolver import FilingResolver
-from app.data.retrieval.hybridrag_retriever import HybridRAGRetriever, RetrievedContext
+from llm.groq_client import GroqClient
+from data.retrieval.filing_resolver import FilingResolver
+from data.retrieval.hybridrag_retriever import HybridRAGRetriever, RetrievedContext
 
 
 @dataclass

{app/llm → llm}/groq_client.py

@@ -2,7 +2,7 @@ from openai import OpenAI
 from typing import List, Dict, Any, Optional, Iterator
 
 from groq import Groq
-from app.config import settings
+from config import settings
 
 
 class GroqClient:

utils/embedding_utils.py

@@ -0,0 +1,31 @@
+from openai import OpenAI
+from config import settings
+
+
+class BGEM3Embedder:
+    """Wraps the NVIDIA serverless BGE-M3 endpoint."""
+    MODEL = "baai/bge-m3"
+
+    def __init__(self):
+        self.client = OpenAI(
+            api_key=settings.NVIDIA_NIM_API,
+            base_url="https://integrate.api.nvidia.com/v1",
+        )
+
+    def embed(self, text: str) -> List[float]:
+        response = self.client.embeddings.create(
+            input=[text],
+            model=self.MODEL,
+            encoding_format="float",
+            extra_body={"truncate": "END"},   # truncate instead of error on long text
+        )
+        return response.data[0].embedding
+
+    def embed_many(self, texts: List[str]) -> List[List[float]]:
+        response = self.client.embeddings.create(
+            input=texts,
+            model=self.MODEL,
+            encoding_format="float",
+            extra_body={"truncate": "END"},
+        )
+        return [d.embedding for d in response.data]

utils/reranker_utils.py

@@ -0,0 +1,58 @@
+import requests
+from typing import List, Dict, Any
+from config import settings
+
+
+class NimReranker:
+    """Nvidia NIM Reranker."""
+    
+    MODEL = "nv-rerank-qa-mistral-4b:1"
+    INVOKE_URL = "https://ai.api.nvidia.com/v1/retrieval/nvidia/reranking"
+
+    def __init__(self):
+        self.session = requests.Session()
+        self.headers = {
+            "Authorization": f"Bearer {settings.NVIDIA_NIM_API}",
+            "Accept": "application/json",
+        }
+
+    def rerank_run(self, query: str, passages: List[str]) -> List[Dict[str, Any]]:
+        """
+        Rerank a list of passages for a given query using Nvidia NIM.
+        
+        Args:
+            query: The question or query string.
+            passages: A list of chunk strings to rerank.
+            
+        Returns:
+            A list of dictionaries containing the text, the ranking score (logit), 
+            and the original index, sorted by score in descending order.
+        """
+        if not passages:
+            return []
+
+        payload = {
+            "model": self.MODEL,
+            "query": {"text": query},
+            "passages": [{"text": p} for p in passages]
+        }
+
+        response = self.session.post(self.INVOKE_URL, headers=self.headers, json=payload)
+        response.raise_for_status()
+        
+        data = response.json()
+        rankings = data.get("rankings", [])
+        
+        results = []
+        for item in rankings:
+            idx = item["index"]
+            results.append({
+                "text": passages[idx],
+                "score": item["logit"],
+                "index": idx
+            })
+            
+        # Sort by score descending
+        results.sort(key=lambda x: x["score"], reverse=True)
+            
+        return results