feat(ci): add ci workflows

.github/fixtures/QandA.jsonl

@@ -0,0 +1,2 @@
+{"prompt": "This is a synthetic fixture FAQ question used only for CI smoke testing.", "completion": "This is a synthetic fixture answer, not real data."}
+{"prompt": "A second synthetic fixture question, distinct wording.", "completion": "A second synthetic fixture answer."}

.github/fixtures/finalcases.json

@@ -0,0 +1,32 @@
+[
+  {
+    "case_id": "fixture_001",
+    "case_title": "CI Fixture Case One",
+    "case_description": "Synthetic fixture case used only to smoke-test the retrieval pipeline in CI. Not real case data.",
+    "case_link": "https://example.com/fixture-1",
+    "sections": [
+      {
+        "section_id": "section_001",
+        "section_title": "Fixture Section",
+        "section_description": "Synthetic section text used only for CI smoke testing."
+      }
+    ],
+    "strong_points": ["Fixture strong point."],
+    "weak_points": ["Fixture weak point."]
+  },
+  {
+    "case_id": "fixture_002",
+    "case_title": "CI Fixture Case Two",
+    "case_description": "A second synthetic fixture case, distinct wording, still only for CI smoke testing.",
+    "case_link": "https://example.com/fixture-2",
+    "sections": [
+      {
+        "section_id": "section_001",
+        "section_title": "Another Fixture Section",
+        "section_description": "More synthetic section text for CI smoke testing."
+      }
+    ],
+    "strong_points": ["Another fixture strong point."],
+    "weak_points": ["Another fixture weak point."]
+  }
+]

.github/workflows/ci.yml

@@ -0,0 +1,40 @@
+name: CI
+
+on:
+  pull_request:
+  push:
+    branches: [main]
+
+jobs:
+  pre-commit:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+      - uses: actions/setup-python@v5
+        with:
+          python-version: "3.12"
+      - run: pip install pre-commit
+      - run: pre-commit run --all-files
+
+  retrieval-smoke-test:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+      - uses: actions/setup-python@v5
+        with:
+          python-version: "3.12"
+          cache: "pip"
+          cache-dependency-path: legalis_api/requirements.txt
+      - run: pip install -r legalis_api/requirements.txt
+      # Real case/FAQ data is intentionally not published (see NOTICE.md) and
+      # is not available in CI. This uses tiny synthetic fixture data to prove
+      # the retrieval pipeline still imports, starts, and returns
+      # correctly-shaped results -- it does NOT measure retrieval quality.
+      # Quality is measured locally against the real dataset via
+      # eval/run_eval.py; see eval/eval_results_*.json for those numbers.
+      - name: Stage fixture data
+        run: |
+          mkdir -p Data
+          cp .github/fixtures/finalcases.json Data/finalcases.json
+          cp .github/fixtures/QandA.jsonl Data/QandA.jsonl
+      - run: python eval/smoke_test.py

.gitignore

@@ -1,11 +1,7 @@
-cases.json
-legal_cases.json
 __pycache__
-legalis_model
 results
-finalcases.json
-trainer.py
-QandA.jsonl
-faq_model
 Data
-Frontend/legalis_ui/node_modules
+/legalis_model
+/faq_model
+/trainer.py
+.venv

eval/smoke_test.py

@@ -0,0 +1,59 @@
+"""
+CI smoke test for the retrieval pipeline.
+
+Unlike run_eval.py, this does NOT measure retrieval quality (that requires
+the real dataset, which is intentionally not published/available in CI). It
+only proves legalis_api/main.py imports, starts, and returns correctly-shaped
+results against tiny synthetic fixture data — catching import errors, crashes,
+and schema breakage on every PR without exposing real data.
+
+Usage:
+    python eval/smoke_test.py
+"""
+
+import os
+import sys
+from pathlib import Path
+
+REPO_ROOT = Path(__file__).resolve().parent.parent
+
+
+def check(condition, message):
+    if not condition:
+        print(f"FAIL: {message}")
+        sys.exit(1)
+
+
+def main():
+    os.chdir(REPO_ROOT / "legalis_api")
+    sys.path.insert(0, str(REPO_ROOT / "legalis_api"))
+    import main as api  # noqa: E402  (import after path/cwd setup, deliberately)
+
+    check(len(api.cases_data) > 0, "no fixture case data loaded")
+    check(len(api.faq_data) > 0, "no fixture FAQ data loaded")
+
+    case_results = api.find_relevant_cases("fixture test query", num_results=2)
+    check(len(case_results) > 0, "find_relevant_cases returned no results")
+    for key in (
+        "case_id",
+        "case_title",
+        "case_link",
+        "similarity_score",
+        "sections",
+        "strong_points",
+        "weak_points",
+    ):
+        check(key in case_results[0], f"case result missing expected key: {key}")
+
+    faq_results = api.find_relevant_faq("fixture test query", num_results=2)
+    check(len(faq_results) > 0, "find_relevant_faq returned no results")
+    for key in ("faq_prompt", "faq_completion", "similarity_score"):
+        check(key in faq_results[0], f"FAQ result missing expected key: {key}")
+
+    print(
+        "Smoke test passed: retrieval pipeline imports, starts, and returns correctly-shaped results."
+    )
+
+
+if __name__ == "__main__":
+    main()