""" LLM ENGINE - TRACK DESIGN ENGINEER GRADE Thinks like a senior track design engineer """ import os import sys import re from typing import List, Dict, Any, Optional from core.bootstrap import ROOT # noqa: F401 from core.config import ProductionConfig from core.audit_logger import get_logger from core.rag import log_llm_answer logger = get_logger("mcp.llm_engine") import ollama import chromadb from sentence_transformers import SentenceTransformer class LLMEngine: def __init__(self, model_name: str = "qwen2:1.5b"): """ Initialize LLM Engine with Engineering Mindset """ self.model_name = model_name self.embedding_model = SentenceTransformer(ProductionConfig.EMBEDDING_MODEL) # Connect to ChromaDB chroma_path = os.path.join(os.path.dirname(os.path.dirname(os.path.abspath(__file__))), "knowledge", "chroma_db") self.chroma_client = chromadb.PersistentClient(path=chroma_path) self.collection = None try: self.collection = self.chroma_client.get_collection(ProductionConfig.CHROMA_COLLECTION) logger.info("Connected to knowledge base") except Exception as e: logger.warning("Knowledge base not found: %s", e) self._check_ollama() def _check_ollama(self): """Check if Ollama is running and model is available""" try: ollama.list() logger.info("Ollama is running") models = ollama.list() model_names = [m['model'] for m in models['models']] if self.model_name not in model_names: logger.warning("Model '%s' not found.", self.model_name) print(f" Available: {', '.join(model_names[:5]) if model_names else 'None'}") print(f" To download: ollama pull {self.model_name}") else: print(f"✅ Model '{self.model_name}' is available") except Exception as e: print(f"❌ Ollama not running!") print(f" Start Ollama with: ollama serve (in a separate window)") print(f" Error: {e}") def search_knowledge(self, question: str, top_k: int = 5) -> List[Dict]: """Search knowledge base for relevant chunks""" if not self.collection: print("⚠️ No knowledge base connection") return [] try: question_embedding = self.embedding_model.encode([question]) results = self.collection.query( query_embeddings=question_embedding.tolist(), n_results=top_k ) formatted = [] if results['documents'] and results['documents'][0]: for i in range(len(results['documents'][0])): formatted.append({ "content": results['documents'][0][i], "metadata": results['metadatas'][0][i], "score": 1 - results['distances'][0][i] if results['distances'] else 0 }) return formatted except Exception as e: print(f"⚠️ Search error: {e}") return [] def build_context(self, results: List[Dict]) -> str: """Build context from search results""" if not results: return "No relevant information found." context_parts = [] for i, result in enumerate(results, 1): source = result['metadata'].get('source', 'Unknown') section = result['metadata'].get('title', 'Unknown Section') content = result['content'][:3000] context_parts.append(f""" [REFERENCE {i}] Source: {source} Section: {section} Content: {content} """) return "\n".join(context_parts) # ============================================================ # ENGINEERING MINDSET - THE KEY UPGRADE # ============================================================ def _get_system_prompt(self) -> str: from core.config import ProductionConfig return ProductionConfig.NR_ASSISTANT_SYSTEM_PROMPT + """ YOUR EXPERTISE: - Network Rail Standards: NR/L2/TRK/2500, NR/L2/TRK/2102, NR/L2/TRK/2049 - Track Design Phases 1-5 (Survey, Design, CAD, Assurance, S&C) YOUR ANSWER FORMAT: 1. REQUIREMENT (cite file + page + clause) 2. REASON 3. APPLICATION 4. RISK/WARNING 5. ACTION """ def _build_prompt(self, question: str, context: str) -> str: """Build the user prompt with engineering focus""" return f""" ENGINEER'S QUESTION: {question} RELEVANT STANDARDS AND KNOWLEDGE: {context} INSTRUCTIONS TO THE SENIOR ENGINEER: 1. Use your 15+ years of track design experience 2. Structure your answer as: REQUIREMENT → REASON → APPLICATION → RISK → ACTION 3. Cite specific standards (e.g., "NR 2102 Section 4.2.1") 4. Give practical, actionable guidance 5. Include any critical warnings or checks 6. If information is missing, say so and suggest where to find it YOUR ENGINEERING ANSWER: """ # ============================================================ # MAIN ASK FUNCTION # ============================================================ def ask(self, question: str, top_k: int = 5) -> Dict: """Ask a question and get an engineering-grade answer""" # 1. Search knowledge base results = self.search_knowledge(question, top_k) if not results: return { "answer": """ ⚠️ I don't have enough information in my knowledge base. RECOMMENDATION: 1. Check if you have the relevant standard in: knowledge/standards/ 2. Run the knowledge loader: python scripts/ingest_knowledge.py 3. Verify the standard PDF is in the correct folder Please ensure your knowledge base contains the relevant standards. """, "sources": [], "confidence": 0, "total_chunks_found": 0 } # 2. Build context context = self.build_context(results) # 3. Build engineering prompt prompt = self._build_prompt(question, context) # 4. Get LLM response try: response = ollama.chat( model=self.model_name, messages=[ {"role": "system", "content": self._get_system_prompt()}, {"role": "user", "content": prompt} ] ) answer = response['message']['content'] if results: meta0 = results[0].get("metadata") or {} log_llm_answer( source_file=str(meta0.get("source", "chroma")), standard_clause=str(meta0.get("clause") or meta0.get("section") or "unknown"), ) except Exception as e: answer = f"❌ Error generating response: {e}" # 5. Extract sources sources = [] for result in results: sources.append({ "source": result['metadata'].get('source', 'Unknown'), "section": result['metadata'].get('title', 'Unknown Section'), "type": result['metadata'].get('type', 'Unknown'), "score": round(result['score'], 3) }) return { "answer": answer, "sources": sources, "confidence": round(results[0]['score'], 3) if results else 0, "total_chunks_found": len(results) } # ============================================================ # DISPLAY FUNCTIONS # ============================================================ def ask_with_sources(self, question: str, top_k: int = 5) -> str: """Get engineering answer with sources""" result = self.ask(question, top_k) output = f""" {'='*65} 🛤️ TRACK DESIGN ENGINEER {'='*65} 📝 QUESTION: {question} {'='*65} 💡 ENGINEERING ANSWER: {'='*65} {result['answer']} {'='*65} 📚 REFERENCES: {'='*65} """ if result['sources']: for i, source in enumerate(result['sources'], 1): output += f""" [{i}] {source['source']} Section: {source['section']} Type: {source['type']} Confidence: {source['score']:.2%} """ else: output += "\n No references available" output += f""" {'='*65} 📊 ASSESSMENT: Confidence: {result['confidence']:.2%} Sources Used: {result['total_chunks_found']} 💡 ENGINEERING JUDGEMENT: {self._get_judgement(result)} {'='*65} """ return output def _get_judgement(self, result: Dict) -> str: """Add engineering judgement to the response""" if result['confidence'] > 0.8: return "✅ High confidence - This information is well supported by standards." elif result['confidence'] > 0.6: return "🟡 Moderate confidence - Verify against the actual standard." elif result['confidence'] > 0.4: return "🔍 Low confidence - Cross-check with the original standard document." else: return "⚠️ Very low confidence - Consider checking the knowledge base content." # ============================================================ # COMMAND LINE INTERFACE # ============================================================ def main(): """Command line interface for Engineering LLM""" print("=" * 65) print(" 🛤️ TRACK DESIGN ENGINEER - AI ASSISTANT") print(" Senior Engineer with 15+ years experience") print("=" * 65) print() engine = LLMEngine() print("\n💡 Ask me anything about UK railway track design.") print(" I'll answer like a senior track design engineer.\n") print("📋 EXAMPLE QUESTIONS:") print(" - What is the ballast depth requirement?") print(" - How to design a ballast ramp?") print(" - What is S&C placement philosophy?") print(" - What should a GAD drawing show?") print(" - Explain stressing of CWR") print() print("-" * 65) while True: try: question = input("\n🔍 Engineer: ").strip() if question.lower() in ['exit', 'quit', 'q']: print("\n👋 Goodbye, Engineer!") break if not question: continue print("\n⏳ Thinking like a Senior Engineer...\n") response = engine.ask_with_sources(question) print(response) except KeyboardInterrupt: print("\n\n👋 Goodbye, Engineer!") break except Exception as e: print(f"❌ Error: {e}") if __name__ == "__main__": main()