RAG Using zvec Vector Datastore and Local Model

1 def chunk_text(text, chunk_size=500, overlap=50):
2     """Split text into overlapping chunks."""
3     chunks = []
4     start = 0
5     while start < len(text):
6         end = start + chunk_size
7         chunks.append(text[start:end])
8         start = end - overlap
9     return chunks

1     schema = zvec.CollectionSchema(
2         name="example",
3         vectors=zvec.VectorSchema("embedding", zvec.DataType.VECTOR_FP32, 768),
4         fields=zvec.FieldSchema("text", zvec.DataType.STRING),
5     )

 1 def build_index():
 2     """Index all text files from the data directory into zvec."""
 3     # Define collection schema (embeddinggemma: 768 dimensions)
 4     schema = zvec.CollectionSchema(
 5         name="example",
 6         vectors=zvec.VectorSchema("embedding", zvec.DataType.VECTOR_FP32, 768),
 7         fields=zvec.FieldSchema("text", zvec.DataType.STRING),
 8     )
 9 
10     db_path = "./zvec_example"
11     if os.path.exists(db_path):
12         import shutil
13         shutil.rmtree(db_path)
14 
15     collection = zvec.create_and_open(path=db_path, schema=schema)
16 
17     docs = []
18     doc_count = 0
19     for root, _, files in os.walk(config["data_dir"]):
20         for file in files:
21             if file.lower().endswith(config["extensions"]):
22                 try:
23                     file_path = Path(root) / file
24                     with open(file_path, "r", encoding="utf-8") as f:
25                         content = f.read()
26                     chunks = chunk_text(content)
27                     for i, chunk in enumerate(chunks):
28                         embedding = get_embedding(chunk)
29                         docs.append(zvec.Doc(
30                             id=f"{file}_{i}",
31                             vectors={"embedding": embedding},
32                             fields={"text": chunk},
33                         ))
34                     doc_count += len(chunks)
35                 except Exception as e:
36                     pass
37 
38     if docs:
39         collection.insert(docs)
40     print(f"Indexed {doc_count} chunks from {config['data_dir']}")
41     return collection

 1 def search_multi_queries(collection, queries, topk=3):
 2     """Search the zvec collection for multiple queries, aggregating and deduplicating chunks."""
 3     all_chunks = []
 4     seen = set()
 5     for query in queries:
 6         chunks = search(collection, query, topk=topk)
 7         for chunk in chunks:
 8             cleaned = chunk.strip()
 9             if cleaned and cleaned not in seen:
10                 seen.add(cleaned)
11                 all_chunks.append(chunk)
12     return all_chunks

 1 def call_llm(system_prompt: str, user_prompt: str, json_format: bool = False) -> str:
 2     """Helper to send a prompt to the Ollama chat model, optionally enforcing JSON format."""
 3     url = f"{OLLAMA_BASE}/api/chat"
 4     payload = {
 5         "model": config["chat_model"],
 6         "stream": False,
 7         "messages": [
 8             {"role": "system", "content": system_prompt},
 9             {"role": "user", "content": user_prompt},
10         ],
11         "options": {
12             "temperature": 0.1,  # Low temperature for deterministic behavior
13         }
14     }
15     if json_format:
16         payload["format"] = "json"
17         
18     req = _make_request(url, payload)
19     try:
20         with urllib.request.urlopen(req) as res:
21             body = json.loads(res.read().decode("utf-8"))
22             return body["message"]["content"]
23     except Exception as e:
24         print(f"Error calling Ollama chat: {e}")
25         return ""

 1 def plan_and_rewrite(question: str) -> dict:
 2     """Planner Agent: Analyzes the question, generates a plan and search queries."""
 3     system_prompt = (
 4         "You are a Plan and Query Rewriter agent. Your task is to analyze the user's question, "
 5         "create a brief search plan, and generate 1 to 3 distinct search queries to retrieve relevant "
 6         "information from a vector database.\n"
 7         "You must respond ONLY with a JSON object in this format:\n"
 8         "{\n"
 9         '  "plan": "brief explanation of what to search for",\n'
10         '  "queries": ["query 1", "query 2"]\n'
11         "}\n"
12         "Do not include any other text."
13     )
14     user_prompt = f"Question: {question}"
15     res_text = call_llm(system_prompt, user_prompt, json_format=True)
16     res_json = parse_json_response(res_text)
17     
18     if not res_json or "queries" not in res_json:
19         res_json = {
20             "plan": f"Direct search for: '{question}'",
21             "queries": [question]
22         }
23     return res_json

 1 def evaluate_context(question: str, snippets: list) -> dict:
 2     """Sufficient Context Agent: Evaluates whether the retrieved snippets contain enough info."""
 3     context_str = "\n\n---\n\n".join(snippets)
 4     system_prompt = (
 5         "You are a Sufficient Context Agent. Your job is to review the user's question, "
 6         "the retrieved context snippets, and determine if the snippets contain all the "
 7         "necessary information to answer the question fully.\n"
 8         "First, mentally draft a potential answer. Then assess if any parts of the question "
 9         "are unanswered or if any crucial information is missing.\n"
10         "You must respond ONLY with a JSON object in this format:\n"
11         "{\n"
12         '  "is_sufficient": true or false (boolean),\n'
13         '  "draft_answer": "a rough draft answer based on the current context",\n'
14         '  "reason": "explanation of what is present or what is missing from the snippets",\n'
15         '  "feedback": "if is_sufficient is false, detailed feedback of what specific keywords, topics, or facts are missing and should be searched for next. If is_sufficient is true, leave this empty."\n'
16         "}\n"
17         "Do not include any other text."
18     )
19     user_prompt = (
20         f"Question: {question}\n\n"
21         f"Retrieved Snippets:\n{context_str}"
22     )
23     res_text = call_llm(system_prompt, user_prompt, json_format=True)
24     res_json = parse_json_response(res_text)
25     
26     if not res_json or "is_sufficient" not in res_json:
27         res_json = {
28             "is_sufficient": True,
29             "draft_answer": "No draft available.",
30             "reason": "Failed to parse sufficiency evaluation, defaulting to sufficient.",
31             "feedback": ""
32         }
33     return res_json

 1 def synthesize_answer(question: str, snippets: list, is_fully_sufficient: bool, sufficiency_reason: str) -> str:
 2     """Synthesis Agent: Generates final grounded response using retrieved context."""
 3     context_str = "\n\n---\n\n".join(snippets)
 4     if is_fully_sufficient:
 5         system_prompt = (
 6             "You are a Synthesis Agent. Write a clear, comprehensive, and accurate final answer "
 7             "to the user's question using ONLY the provided context. Do not extrapolate or assume facts.\n"
 8             f"Context:\n{context_str}"
 9         )
10         user_prompt = f"Question: {question}"
11     else:
12         system_prompt = (
13             "You are a Synthesis Agent. The retrieved context was NOT fully sufficient to answer the question. "
14             "Answer what you can from the provided context, and clearly note what information is missing "
15             "or could not be retrieved from the database. Do not make up any information.\n"
16             f"Reason for insufficiency: {sufficiency_reason}\n\n"
17             f"Context:\n{context_str}"
18         )
19         user_prompt = f"Question: {question}"
20         
21     return call_llm(system_prompt, user_prompt, json_format=False)

 1 $ export MODEL=qwen3:1.7b
 2  $ uv run app.py
 3 Building zvec index from text files …
 4 Indexed 9 chunks from ../data
 5 
 6 Agentic RAG chat ready  (model: nemotron-3-nano:4b)
 7 Type your question, or 'quit' to exit.
 8 
 9 You> What are the main schools of economic thought?
10 
11 🧠 [Planner] Analyzing question and generating search plan...
12    ↳ Plan: Search for a comprehensive list and description of the main schools of economic thought, such as Keynesian, Neoclassical, Marxist, Austrian, etc.
13    ↳ Initial Queries: ['main schools of economic thought', 'major schools of economics overview', 'schools of economic thought']
14 🔍 [Retriever] Searching vector store for queries...
15    ↳ Found 4 unique context snippet(s).
16 🤖 [Sufficiency Check] Evaluating context sufficiency (Iteration 1)...
17    ↳ Sufficiency: False
18    ↳ Reason: Only the Austrian School is described in the snippets; other major schools such as Keynesian, Neoclassical, Marxist, etc., are not mentioned.
19 🔄 [Rewriter] Context insufficient. Feedback: 'Missing information on Keynesian economics, neoclassical theory, Marxist economics, and possibly other contemporary schools. These should be searched for.'
20    Generating new queries based on feedback...
21    ↳ New queries: ['Keynesian economics overview', 'Neoclassical economic theory summary']
22 🔍 [Retriever] Retrieving additional context...
23    ↳ Found 2 new unique snippet(s). Total unique snippets: 6.
24 🤖 [Sufficiency Check] Evaluating context sufficiency (Iteration 2)...
25    ↳ Sufficiency: False
26    ↳ Reason: The snippets only describe the Austrian School in detail. They do not mention other significant schools of economic thought like Keynesian, Neoclassical, or Marxist, which are essential to a complete answer.
27 🔄 [Rewriter] Context insufficient. Feedback: 'Missing keywords: Keynesian, Neoclassical, Marxist, and possibly others such as Institutional economics.'
28    Generating new queries based on feedback...
29    ↳ New queries: ['Marxist economic theory summary', 'Institutional economics schools of thought']
30 🔍 [Retriever] Retrieving additional context...
31    ↳ Found 0 new unique snippet(s). Total unique snippets: 6.
32 🤖 [Sufficiency Check] Evaluating context sufficiency (Iteration 3)...
33    ↳ Sufficiency: False
34    ↳ Reason: The snippets only describe the Austrian School in detail. They do not mention or provide information about other significant schools of economic thought (e.g., Keynesian, Neoclassical, Marxist).
35 ⚠️  [Iteration Limit] Reached max iterations (3). Proceeding to synthesis with partial context.
36 ✍️  [Synthesis] Generating final response...
37 
38 Assistant> **Answer based on the provided context**
39 
40 - The **Austrian School** (also called the Vienna or Psychological School) is described in detail as one school of economic thought that emphasizes the spontaneous organizing power of the price mechanism, advocates a “laissez‑faire” approach, and stresses voluntary contractual agreements with minimal government intervention.
41 
42 **What is missing / could not be retrieved**
43 
44 - The context does **not** mention or provide any information about other significant schools of economic thought such as Keynesian economics, Neoclassical economics, Marxist economics, etc.  
45 - Therefore, I cannot list those schools or describe their characteristics from the given material.
46 
47 **Conclusion**
48 
49 From the supplied text, the only school explicitly described is the **Austrian School**. The existence and description of other major schools (e.g., Keynesian, Neoclassical, Marxist) are not present in the retrieved context.
50 
51 You> who says that Economics is bullshit?
52 
53 🧠 [Planner] Analyzing question and generating search plan...
54    ↳ Plan: Search for statements where someone calls Economics 'bullshit' and identify the speaker or source.
55    ↳ Initial Queries: ['economics is bullshit quote', 'who said economics is bullshit', 'criticism of economics bullshit']
56 🔍 [Retriever] Searching vector store for queries...
57    ↳ Found 3 unique context snippet(s).
58 🤖 [Sufficiency Check] Evaluating context sufficiency (Iteration 1)...
59    ↳ Sufficiency: True
60    ↳ Reason: The snippet explicitly states that Pauli Blendergast, who teaches at the University of Krampton Ohio and is famous for saying economics is bullshit, is the person who makes this claim.
61 ✅ [Sufficiency Check] Context is fully sufficient!
62 ✍️  [Synthesis] Generating final response...
63 
64 Assistant> Pauli Blendergast, an economist who teaches at the University of Krampton, Ohio, is said to claim that “economics is bullshit.”
65 
66 You> ^D
67 Goodbye!