v0.0.6 버전 업데이트

[Gyuho-Han]

Summary by CodeRabbit

Release Notes

• New Features

  • Added Redis-based answer caching for improved response speed with 7-day expiration
  • Added cache management interface to view cached answers and clear cache
  • Implemented graceful degradation—app continues functioning if Redis is unavailable

• Documentation

  • Expanded setup documentation with Redis configuration and Docker deployment instructions

[coderabbitai]

Caution

Review failed

The pull request is closed.

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro

Run ID: cb2da9a9-eb47-4b7a-a8c0-5778e59193fd

📥 Commits

Reviewing files that changed from the base of the PR and between 9d8267b and 212e708.

⛔ Files ignored due to path filters (3)

• cleaned_creation_science_data.csv is excluded by !**/*.csv
• cleaned_dataset.csv is excluded by !**/*.csv
• creation_science_data.csv is excluded by !**/*.csv

📒 Files selected for processing (11)

• README.md
• VERSIONS.md
• app.py
• cache.py
• chain.py
• config.py
• data_pipeline.py
• docker-compose.yml
• ingest_vector_db.py
• requirements.txt
• retriever.py

---

📝 Walkthrough

Walkthrough

Version 0.0.6 adds Redis-backed answer caching with graceful degradation, migrates embeddings from Ollama to Google Gemini, extends the LLM prompt with a CoT reasoning block and a 16384-token context window, refactors the Streamlit app into page-routed functions (render_chat, render_cache_view), simplifies the data pipeline, and adds docker-compose.yml for Redis.

Changes

v0.0.6: Redis Caching, Gemini Embeddings, Prompt/LLM Updates, App Refactor

Layer / File(s)	Summary
Config, dependencies, and Docker infrastructure config.py, requirements.txt, docker-compose.yml	config.py gains load_dotenv(), GEMINI_API_KEY, Redis constants (REDIS_URL, CACHE_ENABLED, CACHE_TTL), Gemini embedding constants, updated chunking params, and drops RAW_CSV_PATH. requirements.txt adds langchain-google-genai, python-dotenv, and redis. docker-compose.yml defines a Redis service with persistence, health check, and restart policy.
Redis cache module cache.py	New module implementing key normalization (_normalize, _make_key via SHA-256), get_cached_answer, set_cached_answer (stores question/answer/created_at hash with TTL), list_cached_answers (scan-based, sorted by recency), clear_cache, and cache_available; all operations fail silently when Redis is unavailable.
Gemini embedding migration retriever.py, ingest_vector_db.py	Both files replace OllamaEmbeddings with GoogleGenerativeAIEmbeddings, add GEMINI_API_KEY presence guards, and construct Chroma via chromadb.PersistentClient. retriever.py additionally adds a 3-attempt retry loop with incremental sleep.
LLM prompt and context window chain.py	Rewrites PROMPT_TEMPLATE with a stricter no-URL directive, a two-sentence conclusion + titled bullet output structure, and an explicit CoT step-by-step block. Introduces LLM_NUM_CTX = 16384 wired into ChatOllama via model_kwargs.
App refactor: page routing and cache integration app.py	Adds cache module imports and a sidebar page selector with cache management UI. render_chat() encapsulates vectorstore init, serializes chat_history_str, short-circuits via get_cached_answer, performs reranking with fallback, streams generation via extract_think_and_answer, assembles final_answer_with_refs, and conditionally stores via set_cached_answer. New render_cache_view() lists cached entries with expanders.
Data pipeline simplification data_pipeline.py	clean_data default input_file changes from RAW_CSV_PATH to CLEANED_CSV_PATH; main() removes the clean_data() call and its result check, calling preprocess_for_rag() directly.
Documentation README.md, VERSIONS.md	README.md adds Redis to prerequisites, documents the caching section and .env variables, and updates the file structure list. VERSIONS.md prepends a v0.0.6 entry covering all feature and configuration changes.

Sequence Diagram(s)

sequenceDiagram
  participant User
  participant render_chat
  participant cache.py
  participant Retriever
  participant generation_chain

  rect rgba(135, 206, 235, 0.5)
    note over render_chat,cache.py: Cache lookup
    User->>render_chat: submit query
    render_chat->>render_chat: serialize chat_history_str
    render_chat->>cache.py: get_cached_answer(query, chat_history_str)
  end

  alt Cache hit
    cache.py-->>render_chat: cached answer
    render_chat-->>User: display cached answer (no LLM call)
  else Cache miss
    rect rgba(255, 165, 0, 0.5)
      note over render_chat,generation_chain: RAG + streaming generation
      render_chat->>Retriever: invoke(query) [reranker or base fallback]
      Retriever-->>render_chat: docs + refs_str
      render_chat->>generation_chain: stream(context, query, chat_history)
      generation_chain-->>render_chat: streamed chunks
      render_chat->>render_chat: extract_think_and_answer → final_answer_with_refs
      render_chat-->>User: display final answer
    end
    rect rgba(144, 238, 144, 0.5)
      note over render_chat,cache.py: Conditional cache store
      alt reranking did not fail
        render_chat->>cache.py: set_cached_answer(query, final_answer_with_refs, chat_history_str)
        cache.py->>cache.py: _normalize + SHA-256 key
        cache.py-->>render_chat: stored with CACHE_TTL
      end
    end
  end

Loading

Estimated code review effort

🎯 4 (Complex) | ⏱️ ~60 minutes

Possibly related PRs

• Gyuho-Han/chatdds#3: Introduced extract_think_and_answer in app.py's streaming flow, which the current PR continues to rely on inside render_chat().
• Gyuho-Han/chatdds#6: Established the refs_str / final_answer_with_refs assembly pattern in app.py that this PR extends by caching the assembled result.
• Gyuho-Han/chatdds#7: The v0.0.5 baseline that last modified chain.py's PROMPT_TEMPLATE, init_generation_chain(), and retriever.py's embedding setup—all of which are further changed in this PR.

Poem

🐇 Hop, hop! A cache appears,
SHA-256 keys, no duplicate fears.
Gemini embeddings shine so bright,
CoT reasoning guides the light.
Redis stores each answer neat—
This bunny's work is quite complete! ✨

✨ Finishing Touches

📝 Generate docstrings

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch v0.0.6

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}