fix: ragas evaluator - column mapping, embeddings param, venv setup

AGENTS.md

@@ -0,0 +1,36 @@
+=== AGENTS.md ===
+
+# ACG Protocol
+
+Audited Context Generation (ACG) Protocol — a dual-layer standard for veracity assurance in RAG systems. Python-based project for grounding facts and verifying reasoning.
+
+## Directory Structure
+
+  README.md
+  Usage.md
+  requirements.txt
+  AGENTS.md
+  docs/
+    ACG_PROTOCOL.md       — Core ACG protocol specification
+    UGVP_PROTOCOL.md       — Universal Grounding & Verification Protocol (Layer 1)
+    RSVP_PROTOCOL.md       — Reasoning & Synthesis Verification Protocol (Layer 2)
+  src/
+    agent.py               — Main agent entry point / inference
+    config.py              — Configuration
+    evaluation_data.py     — Evaluation data helpers
+    indexer.py             — Chunk + SHI indexer for URLs
+    mongodb_client.py      — MongoDB client for store/retrieve
+    ragas_evaluator.py     — RAGAS evaluation integration
+    ugvp_protocol.py       — UGVP protocol implementation
+    utils.py               — Utility functions
+
+## Stack
+
+Python 3.12+, MongoDB, aiohttp, google-generativeai, RAGAS
+
+## Commands
+
+Setup:    python -m venv venv && source venv/bin/activate && pip install -r requirements.txt
+Index:    python src/indexer.py --sentences_per_chunk 5 <url>
+Agent:    python src/agent.py
+Tests:    pytest (if configured)

requirements.txt

@@ -13,7 +13,7 @@ click==8.3.0
 distro==1.9.0
 dnspython==2.8.0
 docstring_parser==0.17.0
-fastuuid==0.13.5
+fastuuid==0.14.0
 filelock==3.19.1
 frozenlist==1.8.0
 fsspec==2025.9.0
@@ -44,11 +44,11 @@ langchain-community
 langchain-litellm
 langchain-google-vertexai
 langchain-google-genai
-litellm
+litellm==1.85.5
 MarkupSafe==3.0.3
 mcp==1.16.0
 multidict==6.7.0
-openai==2.2.0
+openai==2.20.0
 opentelemetry-api==1.37.0
 opentelemetry-instrumentation==0.58b0
 opentelemetry-instrumentation-threading==0.58b0

src/evaluation_data.py

@@ -41,9 +41,9 @@ async def get_evaluation_data() -> List[Dict[str, Any]]:
         reference_answer = agent_answer 
 
         evaluation_data.append({
-            "question": question,
-            "contexts": verified_contexts,
-            "answer": agent_answer,
+            "user_input": question,
+            "retrieved_contexts": verified_contexts,
+            "response": agent_answer,
             "reference": reference_answer,
         })
 

src/ragas_evaluator.py

@@ -6,7 +6,7 @@
 
 from datasets import Dataset
 from ragas import evaluate
-from ragas.metrics import (
+from ragas.metrics.collections import (
     faithfulness,
     answer_relevancy,
     context_recall,
@@ -25,9 +25,9 @@ async def run_ragas_evaluation(data: List[Dict[str, Any]]):
     Runs RAGAS evaluation on the provided dataset using Google Gemini via ChatGoogleGenerativeAI.
 
     @param data: A list of dictionaries, where each dictionary contains:
-                 'question': The user query.
-                 'contexts': A list of retrieved contexts.
-                 'answer': The generated answer.
+                 'user_input': The user query.
+                 'retrieved_contexts': A list of retrieved contexts.
+                 'response': The generated answer.
                  'reference': The ground truth answer (single string).
     """
     if not data:
@@ -59,7 +59,8 @@ async def run_ragas_evaluation(data: List[Dict[str, Any]]):
                 context_recall,
                 context_precision,
             ],
-            llm=ragas_llm
+            llm=ragas_llm,
+            embeddings=ragas_embeddings,
         )
         log.info("RAGAS evaluation completed.")
         log.info(result.to_pandas().to_json(orient="records", indent=2))