startup

Dockerfile

@@ -7,9 +7,8 @@ RUN pip install --no-cache-dir -r requirements.txt
 
 COPY app.py fiscal.py auth.py plaid_client.py rebuild_index.py ./
 COPY docs/ ./docs/
-COPY chroma_db/ ./chroma_db/
 
 EXPOSE 7860
 
-# Skip rebuild if chroma_db exists, then start
+# Rebuild ChromaDB then start the server
 CMD ["sh", "-c", "python rebuild_index.py && uvicorn app:app --host 0.0.0.0 --port 7860"]

rebuild_index.py

@@ -5,38 +5,35 @@ from langchain_community.vectorstores import Chroma
 import shutil
 import os
 
-CHROMA_DIR = "./chroma_db"
-DOCS_FILE = "./docs/fiscal_knowledge.md"
+# Wipe old index
+if os.path.exists("./chroma_db"):
+    shutil.rmtree("./chroma_db")
+    print("Deleted old chroma_db/")
 
-# Only rebuild if chroma_db doesn't exist or docs changed
-if os.path.exists(CHROMA_DIR):
-    print("chroma_db/ exists, skipping rebuild. Delete chroma_db/ to force rebuild.")
-else:
-    print("Building new index...")
-    
-    loader = TextLoader(DOCS_FILE, encoding="utf-8")
-    docs = loader.load()
-    text = docs[0].page_content
-    print(f"Loaded {len(text):,} characters")
+# Load markdown
+loader = TextLoader("./docs/fiscal_knowledge.md", encoding="utf-8")
+docs = loader.load()
+text = docs[0].page_content
+print(f"Loaded {len(text):,} characters")
 
-    splitter = MarkdownHeaderTextSplitter(
-        headers_to_split_on=[("##", "question")],
-        strip_headers=False,
-    )
-    chunks = splitter.split_text(text)
-    print(f"Split into {len(chunks)} chunks")
+splitter = MarkdownHeaderTextSplitter(
+    headers_to_split_on=[("##", "question")],
+    strip_headers=False,
+)
+chunks = splitter.split_text(text)
 
-    print("Loading embedding model...")
-    embeddings = HuggingFaceEmbeddings(
-        model_name="nomic-ai/nomic-embed-text-v1",
-        model_kwargs={"trust_remote_code": True},
-    )
+print(f"Split into {len(chunks)} chunks")
+print("Loading embedding model (first run downloads ~270MB)...")
+embeddings = HuggingFaceEmbeddings(
+    model_name="nomic-ai/nomic-embed-text-v1",
+    model_kwargs={"trust_remote_code": True},
+)
 
-    vectorstore = Chroma.from_documents(
-        chunks,
-        embedding=embeddings,
-        persist_directory=CHROMA_DIR,
-    )
-    print(f"Indexed {len(chunks)} chunks into chroma_db/")
+vectorstore = Chroma.from_documents(
+    chunks,
+    embedding=embeddings,
+    persist_directory="./chroma_db",
+)
 
-print("Ready.")
+print(f"Indexed {len(chunks)} chunks into chroma_db/")
+print("Done. Ready to deploy.")