Practical notes: I Learned RAG by Building a Q&A System

Use this as an operator-facing rebuild of the ideas in “I Learned RAG by Building a Q&A System”: clear stages, ordered code slots, and recovery notes that survive a handoff. The Overview stage works best when treated as a measurable surface. Capture one golden transcript, one failure case, and the rollback note before expanding scope. Record timings and token or query cost next to functional results. Cost visibility early prevents surprise bills when the path moves from demo to shared environments.

A quick note on RAG

For the A quick note on stage, define the inputs, the owner of the step, and the exit criteria before changing code. Operators should be able to re-run the step from a known checkpoint without guessing hidden state. Keep configuration outside application code. Environment files, secret stores, and feature flags belong in one place operators can audit without reading the whole graph. Cite the passages that actually grounded the answer. Without citations, operators cannot tell hallucination from an indexing gap.

How the system is structured

For the How the system is stage, define the inputs, the owner of the step, and the exit criteria before changing code. Operators should be able to re-run the step from a known checkpoint without guessing hidden state. Document the happy path and the recovery path together. Retries, human gates, and dead-letter handling are part of the product, not later polish. Cite the passages that actually grounded the answer. Without citations, operators cannot tell hallucination from an indexing gap.

Building the ingestion pipeline

For the Building the ingestion pipeline stage, define the inputs, the owner of the step, and the exit criteria before changing code. Operators should be able to re-run the step from a known checkpoint without guessing hidden state. Prefer small, testable units over sprawling scripts. When a step fails, the failure should point at a single responsibility rather than a tangled pipeline. Cite the passages that actually grounded the answer. Without citations, operators cannot tell hallucination from an indexing gap.

Downloading

For the Downloading stage, define the inputs, the owner of the step, and the exit criteria before changing code. Operators should be able to re-run the step from a known checkpoint without guessing hidden state. Treat this stage as a contract between inputs and validated outputs. Name the artifacts, define success checks, and refuse silent partial completion. Cite the passages that actually grounded the answer. Without citations, operators cannot tell hallucination from an indexing gap.

{
  "doc_id": "bsp-mpr-2023-q3",
  "title": "Monetary Policy Report Q3 2023",
  "period_label": "Q3 2023",
  "publication_date": "2023-08-17",
  "url": "https://www.bsp.gov.ph/..."
}

Parsing

For the Parsing stage, define the inputs, the owner of the step, and the exit criteria before changing code. Operators should be able to re-run the step from a known checkpoint without guessing hidden state. Record timings and token or query cost next to functional results. Cost visibility early prevents surprise bills when the path moves from demo to shared environments. Cite the passages that actually grounded the answer. Without citations, operators cannot tell hallucination from an indexing gap.

Chunking, embedding, and upserting

For the Chunking embedding and upserting stage, define the inputs, the owner of the step, and the exit criteria before changing code. Operators should be able to re-run the step from a known checkpoint without guessing hidden state. Keep configuration outside application code. Environment files, secret stores, and feature flags belong in one place operators can audit without reading the whole graph. Cite the passages that actually grounded the answer. Without citations, operators cannot tell hallucination from an indexing gap.

def chunk_text_by_tokens(text, enc, chunk_size=512, overlap=64):
    tokens = enc.encode(text)
    stride = chunk_size - overlap
    chunks = []
    start = 0
    while start < len(tokens):
        window = tokens[start : start + chunk_size]
        chunks.append(enc.decode(window))
        if start + chunk_size >= len(tokens):
            break
        start += stride
    return chunks

The retrieval layer and the recency problem

For the The retrieval layer and stage, define the inputs, the owner of the step, and the exit criteria before changing code. Operators should be able to re-run the step from a known checkpoint without guessing hidden state. Document the happy path and the recovery path together. Retries, human gates, and dead-letter handling are part of the product, not later polish. Cite the passages that actually grounded the answer. Without citations, operators cannot tell hallucination from an indexing gap.

RECENCY_WEIGHT = 0.85

def combined_score(hit):
    relevance = hit.score / max_score
    recency = (pub_date - min_date).days / date_span_days  # 0.0 to 1.0
    return (1 - RECENCY_WEIGHT) * relevance + RECENCY_WEIGHT * recency

# scope to Q1 2023 only
retrieve_chunks(query, period_label_key="q1_2023")

# scope to a date range
retrieve_chunks(query, publication_date_from="2023-01-01", publication_date_to="2023-12-31")

The generation layer

For the The generation layer stage, define the inputs, the owner of the step, and the exit criteria before changing code. Operators should be able to re-run the step from a known checkpoint without guessing hidden state. Prefer small, testable units over sprawling scripts. When a step fails, the failure should point at a single responsibility rather than a tangled pipeline. Cite the passages that actually grounded the answer. Without citations, operators cannot tell hallucination from an indexing gap. For the The generation layer stage, define the inputs, the owner of the step, and the exit criteria before changing code. Operators should be able to re-run the step from a known checkpoint without guessing hidden state. Record timings and token or query cost next to functional results. Cost visibility early prevents surprise bills when the path moves from demo to shared environments.

Problems you actually ran into

When working through the Problems you actually ran stage, write down the contract first: required inputs, success signal, and what happens on partial failure. That checklist keeps later code changes honest. Keep configuration outside application code. Environment files, secret stores, and feature flags belong in one place operators can audit without reading the whole graph. Measure recall on a fixed question set before tuning prompts. Prompt churn rarely fixes a weak retrieval surface.

PDF boilerplate polluting chunks

When working through the PDF boilerplate polluting chunks stage, write down the contract first: required inputs, success signal, and what happens on partial failure. That checklist keeps later code changes honest. Document the happy path and the recovery path together. Retries, human gates, and dead-letter handling are part of the product, not later polish. Measure recall on a fixed question set before tuning prompts. Prompt churn rarely fixes a weak retrieval surface.

def _clean_parsed_text(text):
    text = text.replace("\x0c", "\n\n")
    text = re.sub(r"Classification:\s*GENERAL\s*\n", "", text)
    text = re.sub(r"Monetary Policy Report\s*[--][^\n]+\|\s*\d+\s*\n", "", text)
    text = re.sub(r"\n{3,}", "\n\n", text)
    return text.strip()

Accidental collection wipe

When working through the Accidental collection wipe stage, write down the contract first: required inputs, success signal, and what happens on partial failure. That checklist keeps later code changes honest. Prefer small, testable units over sprawling scripts. When a step fails, the failure should point at a single responsibility rather than a tangled pipeline. Measure recall on a fixed question set before tuning prompts. Prompt churn rarely fixes a weak retrieval surface. When working through the Accidental collection wipe stage, write down the contract first: required inputs, success signal, and what happens on partial failure. That checklist keeps later code changes honest. Record timings and token or query cost next to functional results. Cost visibility early prevents surprise bills when the path moves from demo to shared environments.

def _should_recreate_qdrant_collection():
    if _is_production_environment():
        return False
    explicit = os.environ.get("QDRANT_RECREATE_COLLECTION", "").lower()
    if explicit in ("1", "true", "yes"):
        return True
    app_env = (os.environ.get("ENVIRONMENT") or "").lower()
    return app_env in ("development", "dev", "local")

Duplicate chunks after re-ingestion

The Duplicate chunks after re-ingestion stage works best when treated as a measurable surface. Capture one golden transcript, one failure case, and the rollback note before expanding scope. Keep configuration outside application code. Environment files, secret stores, and feature flags belong in one place operators can audit without reading the whole graph. Separate chunking policy from retrieval policy. Changing one should not force a rewrite of the other when quality metrics move.

client.delete(
    collection_name=collection,
    points_selector=models.Filter(must=[
        models.FieldCondition(
            key="source_file",
            match=models.MatchValue(value=source_file)
        )
    ]),
)

Cold start latency on the free tier

The Cold start latency on stage works best when treated as a measurable surface. Capture one golden transcript, one failure case, and the rollback note before expanding scope. Document the happy path and the recovery path together. Retries, human gates, and dead-letter handling are part of the product, not later polish. Separate chunking policy from retrieval policy. Changing one should not force a rewrite of the other when quality metrics move.

Picking a chunk size

The Picking a chunk size stage works best when treated as a measurable surface. Capture one golden transcript, one failure case, and the rollback note before expanding scope. Prefer small, testable units over sprawling scripts. When a step fails, the failure should point at a single responsibility rather than a tangled pipeline. Separate chunking policy from retrieval policy. Changing one should not force a rewrite of the other when quality metrics move. The Picking a chunk size stage works best when treated as a measurable surface. Capture one golden transcript, one failure case, and the rollback note before expanding scope. Record timings and token or query cost next to functional results. Cost visibility early prevents surprise bills when the path moves from demo to shared environments.

What you’d do differently

For the What you d do stage, define the inputs, the owner of the step, and the exit criteria before changing code. Operators should be able to re-run the step from a known checkpoint without guessing hidden state. Keep configuration outside application code. Environment files, secret stores, and feature flags belong in one place operators can audit without reading the whole graph. Cite the passages that actually grounded the answer. Without citations, operators cannot tell hallucination from an indexing gap.

Try it

For the Try it stage, define the inputs, the owner of the step, and the exit criteria before changing code. Operators should be able to re-run the step from a known checkpoint without guessing hidden state. Document the happy path and the recovery path together. Retries, human gates, and dead-letter handling are part of the product, not later polish. Cite the passages that actually grounded the answer. Without citations, operators cannot tell hallucination from an indexing gap.

Operational checklist

When working through the Operational checklist stage, write down the contract first: required inputs, success signal, and what happens on partial failure. That checklist keeps later code changes honest.

Treat this stage as a contract between inputs and validated outputs. Name the artifacts, define success checks, and refuse silent partial completion.

Measure recall on a fixed question set before tuning prompts. Prompt churn rarely fixes a weak retrieval surface.

Pin dependency versions and record the image digest that ran the demo. Reproducibility beats tribal knowledge.

Record timings and token or query cost next to functional results. Cost visibility early prevents surprise bills when the path moves from demo to shared environments.

Measure recall on a fixed question set before tuning prompts. Prompt churn rarely fixes a weak retrieval surface.

Before promoting the stack, freeze versions, capture a golden transcript for the critical path, and confirm rollback steps. Shared environments need rate limits, tenancy checks, and a clear owner for secret rotation. Prefer boring reliability over clever one-off demos.

Batch note for a30a0175d827: keep provider keys out of the repo, set a per-session token ceiling, and store transcripts next to the eval fixtures so later model swaps stay comparable.