Interview-ready pass: pin deps, fix stranded demo link, real eval numbers, CI

.github/workflows/ci.yml

@@ -0,0 +1,25 @@
+name: CI
+
+on:
+  push:
+  pull_request:
+    branches: [main]
+
+jobs:
+  test:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+
+      - uses: actions/setup-python@v5
+        with:
+          python-version: "3.11"
+
+      - name: Install dependencies
+        run: pip install -r requirements.txt
+
+      - name: Run test suite
+        run: pytest -q
+
+    # Embeddings are local (FastEmbed, BAAI/bge-small-en-v1.5) and the test
+    # suite has no external API dependency, so this needs no secrets.

LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 Raghav Malani
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

README.md

@@ -1,5 +1,11 @@
 # Codebase Intelligence RAG
 
+[![CI](https://github.com/Ragh234/codebase-intelligence-rag/actions/workflows/ci.yml/badge.svg)](https://github.com/Ragh234/codebase-intelligence-rag/actions/workflows/ci.yml)
+![Python](https://img.shields.io/badge/python-3.10%2B-blue)
+[![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](LICENSE)
+
+**Live demo:** https://codebase-intelligence-rag-hqfteuyappwhonm8yje4agk.streamlit.app/
+
 Codebase Intelligence RAG is a focused RAG/ML engineering project for asking grounded questions about GitHub codebases. It is not a full-stack product: Streamlit is only a thin demo layer, and the core work is repository ingestion, code-aware retrieval, reranking, context construction, and grounded generation with exact citations.
 
 ## Why Basic Vector Search Is Not Enough For Code
@@ -174,15 +180,29 @@ Run:
 python -m evaluation.run_evaluation
 ```
 
-The evaluator reports Hit Rate, Recall@K, Precision@K, and MRR for BM25, vector, hybrid, hybrid + graph, and full reranked retrieval.
+The evaluator indexes a small fixture repo (`tests/fixtures/sample_repo`) and reports Hit Rate, Recall@5, Precision@5, and MRR across five retrieval strategies against a 5-question dataset. Measured results:
+
+| Strategy | Hit Rate | Recall@5 | Precision@5 | MRR |
+|---|---|---|---|---|
+| BM25 only | 0.80 | 0.80 | 0.44 | 0.60 |
+| Vector only | 1.00 | 1.00 | 0.52 | 0.717 |
+| Hybrid (RRF) | 1.00 | 1.00 | **0.56** | **0.75** |
+| Hybrid + Graph | 1.00 | 1.00 | **0.56** | 0.717 |
+| Full (Hybrid + Graph + rerank) | 1.00 | 1.00 | 0.52 | 0.717 |
+
+Takeaways, read honestly rather than cherry-picked:
+
+- **Hybrid clearly beats either single-strategy baseline** — BM25-only misses one question entirely (hit rate 0.80 vs. 1.00), and plain vector search has the weakest precision/MRR of anything that hits every question.
+- **Graph expansion and reranking don't show a clear win on this 5-question set.** "Full" (with reranking) actually scores fractionally *lower* on precision and MRR than plain Hybrid here. With only 5 questions, that's within noise, not evidence the reranker hurts — but it's also not evidence it helps. A fair statement is: hybrid retrieval is the load-bearing improvement in this project; graph expansion and reranking are structurally sound (tested independently in `tests/`) but this eval set is too small to say whether they help end-to-end.
+- Scaling `DATASET` in `evaluation/run_evaluation.py` beyond 5 questions would be the natural next step to get a statistically meaningful answer on reranking.
 
 ## Tests
 
 ```bash
 pytest
 ```
 
-Tests cover ingestion, parser extraction, chunk metadata, BM25, vector retrieval, hybrid retrieval, metadata filtering, graph expansion, citations, context construction, LLM request construction, and no-LLM evidence fallback.
+21 tests pass, covering ingestion, parser extraction, chunk metadata, BM25, vector retrieval, hybrid retrieval, metadata filtering, graph expansion, citations, context construction, LLM request construction, and no-LLM evidence fallback.
 
 ## How This Differs From A Basic RAG Chatbot
 
@@ -193,7 +213,7 @@ documents -> text chunks -> embeddings -> vector search -> LLM
 ```
 
 This project:
-Deployed Link - https://codebase-intelligence-rag-hqfteuyappwhonm8yje4agk.streamlit.app/
+
 ```text
 code -> AST/code entities -> metadata-rich chunks -> semantic + lexical retrieval
      -> graph relationships -> hybrid ranking -> bounded context -> grounded generation

requirements.txt

@@ -1,14 +1,14 @@
-streamlit
-gitpython
-tree-sitter
-tree-sitter-python
-tree-sitter-javascript
-tree-sitter-typescript
-tree-sitter-java
-tree-sitter-cpp
-tree-sitter-go
-networkx
-fastembed
-numpy
-httpx
-pytest
+streamlit==1.63.0
+gitpython==3.1.61
+tree-sitter==0.26.0
+tree-sitter-python==0.25.0
+tree-sitter-javascript==0.25.0
+tree-sitter-typescript==0.23.2
+tree-sitter-java==0.23.5
+tree-sitter-cpp==0.23.4
+tree-sitter-go==0.25.0
+networkx==3.6.1
+fastembed==0.8.0
+numpy==2.5.2
+httpx==0.28.1
+pytest==9.1.1