Add RAG service with FastAPI, RAGEngine, config, and deps by NaveenBuidl · Pull Request #3 · NaveenBuidl/evalens

NaveenBuidl · 2026-04-08T07:41:34Z

Motivation

Provide a small Retrieval-Augmented Generation (RAG) service to index a PDF/text corpus and serve queries with optional Groq-powered generation.
Centralize runtime configuration in a YAML file and environment variables so credentials (like GROQ_API_KEY) are loaded from .env and config.yaml.
Add required dependencies and a lightweight API so the RAG functionality can be run as a service.

Description

Add app/config.py to load settings from config.yaml and GROQ_API_KEY from environment via python-dotenv into a Settings dataclass.
Implement app/rag.py containing RAGEngine which uses chromadb persistent client, a SentenceTransformer embedding function, and PyMuPDF to extract text from PDFs, chunk documents, index chunks, and perform similarity queries.
Add app/main.py exposing a FastAPI app with a startup ingest() call and a /query POST endpoint that returns retrieved chunks, sources, and an optional Groq-generated answer when GROQ_API_KEY is set.
Include config.yaml defaults, .env.example with GROQ_API_KEY placeholder, and requirements.txt listing runtime dependencies (fastapi, uvicorn, pymupdf, chromadb, sentence-transformers, groq, python-dotenv, pyyaml).

Testing

No automated tests were executed as part of this rollout.

Codex Task

Copilot

Pull request overview

Adds a minimal FastAPI-based Retrieval-Augmented Generation (RAG) microservice that can ingest a local corpus into a persistent ChromaDB collection and serve similarity queries with optional Groq-based generation, with runtime configuration sourced from config.yaml and environment variables.

Changes:

Introduces RAGEngine for PDF/text ingestion, chunking, embedding, ChromaDB persistence, and query-time retrieval + optional Groq completion.
Adds configuration loading via YAML + .env and exposes a FastAPI app with startup ingestion and a /query endpoint.
Adds initial runtime dependency list, default config.yaml, and .env.example.

Reviewed changes

Copilot reviewed 5 out of 6 changed files in this pull request and generated 9 comments.

Show a summary per file

File	Description
`requirements.txt`	Adds runtime dependencies for FastAPI, ChromaDB, embeddings, PDF parsing, Groq, and config loading.
`config.yaml`	Provides default runtime configuration (corpus path, chunking, retrieval params, model, Chroma persistence).
`app/config.py`	Implements settings loading from YAML + `.env` into a `Settings` dataclass.
`app/rag.py`	Implements ingestion + retrieval and optional Groq answer generation.
`app/main.py`	Exposes the service via FastAPI startup ingestion and `/query` endpoint.
`.env.example`	Documents required Groq API key env var.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Copilot · 2026-04-08T07:45:03Z

@@ -0,0 +1,8 @@
+corpus_path: "D:/Evalens/corpus/intercom_external/raw_pdfs"


corpus_path default is a machine-specific absolute Windows path, which will break on other environments and in CI/containers. Consider making the default a relative path (e.g., ./corpus) and/or sourcing it from an environment variable instead of hardcoding a local drive path.

Suggested change

corpus_path: "D:/Evalens/corpus/intercom_external/raw_pdfs"

corpus_path: "./corpus"

Copilot · 2026-04-08T07:45:04Z

+retrieval_k: 4
+model: "llama-3.1-8b-instant"
+embedding_model: "sentence-transformers/all-MiniLM-L6-v2"
+chroma_path: ".chroma"


Default chroma_path is set to .chroma, but the repository’s ignore patterns currently exclude chroma_db/ rather than .chroma/. This makes it easy to accidentally commit the persistent vector DB; consider aligning the default path with the ignored directory name or updating ignore rules accordingly.

Suggested change

chroma_path: ".chroma"

chroma_path: "chroma_db"

Copilot · 2026-04-08T07:45:04Z

+
+    return Settings(
+        corpus_path=cfg.get("corpus_path", "D:/Evalens/corpus/intercom_external/raw_pdfs"),


load_settings() falls back to a machine-specific absolute Windows corpus_path. If config.yaml is missing/misconfigured in another environment, the service will fail in a non-obvious way. Prefer a portable default (relative path) and/or allow overriding via an env var (e.g., CORPUS_PATH).

Suggested change

return Settings(

corpus_path=cfg.get("corpus_path", "D:/Evalens/corpus/intercom_external/raw_pdfs"),

default_corpus_path = cfg_file.parent / "corpus"

corpus_path = os.getenv("CORPUS_PATH") or cfg.get("corpus_path") or str(default_corpus_path)

return Settings(

corpus_path=corpus_path,

Copilot · 2026-04-08T07:45:04Z

+        existing_count = self._collection.count()
+        if existing_count > 0:
+            return {"indexed_files": 0, "indexed_chunks": existing_count, "skipped": True}


ingest() permanently skips indexing when the collection already has any documents (count() > 0). With a persistent Chroma DB, this means newly added/updated corpus files will never be indexed unless the DB is manually deleted. Consider tracking ingested sources (e.g., by file mtime/hash) and upserting new chunks, or providing an explicit force_reindex/reset option.

Copilot · 2026-04-08T07:45:04Z

+            for idx, chunk in enumerate(chunks):
+                ids.append(f"{source_path.stem}-{idx}")
+                docs.append(chunk)
+                metas.append({"source": str(source_path), "chunk_index": idx})


Chunk IDs are derived from source_path.stem and idx, which can collide when different files share the same stem (e.g., report.pdf and report.txt, or duplicates in different subfolders), causing Chroma add() failures or overwrites. Consider incorporating the full relative path (or a stable hash of it) and the suffix into the ID.

Copilot · 2026-04-08T07:45:05Z

+            for idx, chunk in enumerate(chunks):
+                ids.append(f"{source_path.stem}-{idx}")
+                docs.append(chunk)
+                metas.append({"source": str(source_path), "chunk_index": idx})


source metadata stores the full server filesystem path and is later returned in the API response. This can leak internal directory structure to clients. Consider storing/returning only a safe identifier (e.g., basename, relative path within corpus, or a document ID) and keep absolute paths server-side only.

Suggested change

metas.append({"source": str(source_path), "chunk_index": idx})

metas.append({"source": source_path.name, "chunk_index": idx})

Copilot · 2026-04-08T07:45:05Z

+@app.on_event("startup")
+def startup_event() -> None:
+    rag.ingest()


Running rag.ingest() synchronously during FastAPI startup can block the service from becoming ready for a long time (potentially causing health-check failures/timeouts) when the corpus is large. Consider moving ingestion to a background task, a separate admin endpoint/CLI, or making startup ingestion optional via configuration.

Copilot · 2026-04-08T07:45:05Z

+@app.on_event("startup")
+def startup_event() -> None:
+    rag.ingest()


Ingestion at startup can race in multi-worker deployments (e.g., uvicorn --workers N) where multiple processes call ingest() simultaneously against the same persistent Chroma path, potentially causing duplicate-ID errors or DB corruption. Consider ensuring single-worker ingestion, adding an inter-process lock, or using a dedicated one-off ingestion job.

Copilot · 2026-04-08T07:45:05Z

+    except Exception as e:  # minimal tracer-bullet error handling
+        raise HTTPException(status_code=500, detail=str(e)) from e


Returning detail=str(e) for all unexpected exceptions can leak internal error messages, file paths, and implementation details to clients. Prefer returning a generic 500 message and logging the exception server-side (with a request/correlation ID if possible).

Ingest markdown and text files alongside PDFs

ab9443a

NaveenBuidl added the codex label Apr 8, 2026 — with ChatGPT Codex Connector

Copilot AI review requested due to automatic review settings April 8, 2026 07:41

Copilot started reviewing on behalf of NaveenBuidl April 8, 2026 07:42 View session

Copilot AI reviewed Apr 8, 2026

View reviewed changes

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Add RAG service with FastAPI, RAGEngine, config, and deps#3

Add RAG service with FastAPI, RAGEngine, config, and deps#3
NaveenBuidl wants to merge 1 commit into
mainfrom
codex/build-minimal-rag-app-with-fastapi-ee2e7h

NaveenBuidl commented Apr 8, 2026

Uh oh!

Copilot AI left a comment

Uh oh!

Copilot AI Apr 8, 2026

Uh oh!

Copilot AI Apr 8, 2026

Uh oh!

Copilot AI Apr 8, 2026

Uh oh!

Copilot AI Apr 8, 2026

Uh oh!

Copilot AI Apr 8, 2026

Uh oh!

Copilot AI Apr 8, 2026

Uh oh!

Copilot AI Apr 8, 2026

Uh oh!

Copilot AI Apr 8, 2026

Uh oh!

Copilot AI Apr 8, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

		@@ -0,0 +1,8 @@
		corpus_path: "D:/Evalens/corpus/intercom_external/raw_pdfs"

	corpus_path: "D:/Evalens/corpus/intercom_external/raw_pdfs"
	corpus_path: "./corpus"


		return Settings(
		corpus_path=cfg.get("corpus_path", "D:/Evalens/corpus/intercom_external/raw_pdfs"),

	metas.append({"source": str(source_path), "chunk_index": idx})
	metas.append({"source": source_path.name, "chunk_index": idx})

		except Exception as e: # minimal tracer-bullet error handling
		raise HTTPException(status_code=500, detail=str(e)) from e

Conversation

NaveenBuidl commented Apr 8, 2026

Motivation

Description

Testing

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Copilot AI Apr 8, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 8, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 8, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 8, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 8, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 8, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 8, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 8, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 8, 2026

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants