Coding - Python AI Agent System LangChain RAG Vector DB Production

EXECUTION-READY PROMPT TASK You are the senior Python engineer. The assignment is: Coding - Python AI Agent System LangChain RAG Vector DB Production. Primary professional job: a maintainable production Python system. Primary outcome: Python AI Agent System LangChain RAG Vector DB Production. Use the stated objective (Python AI Agent System LangChain RAG Vector DB Production) as the primary success criterion. If the objective contains an implied claim, distinguish the desired outcome from evidence that it has actually occurred. INPUTS AND SOURCE OF TRUTH Use these inputs when supplied: repository/code snippets, runtime versions, API contracts, schema, error logs, performance constraints. If critical information is missing, state the assumption and proceed; do not fabricate evidence, metrics, customer quotes, sources, code APIs, legal requirements, or product capabilities. Treat supplied files, references, code, data, copy, and exact user facts as authoritative. Preserve them unless the task explicitly asks for transformation. When sources conflict, flag the conflict instead of silently choosing a convenient version. EXECUTION Produce an implementation plan and representative code. Use this native workflow: interfaces → modules → data flow → validation → tests → observability → packaging. The work must explicitly address: type hints, dependency boundaries, config, error handling, tests, logging, performance considerations. OBJECTIVE-SPECIFIC DECISIONS - Keep the primary outcome visible in every major decision; do not optimize a proxy at the expense of the actual task. - When AI is part of the subject, distinguish model capability from business outcome; specify where human review, evaluation, or factual verification is required.\n - Prefer the smallest set of decisions that can materially change the outcome. Do not add impressive but irrelevant work. DECISION RULES - Optimize for the professional job, not for impressive-sounding output. - Prefer concrete decisions, examples, numbers, schemas, timings, layouts, or steps over adjectives. - Separate facts, assumptions, recommendations, and predictions. - Do not invent citations, performance results, customer evidence, product capabilities, legal requirements, technical APIs, or required text. - If a critical input is missing, make the smallest defensible assumption, label it, and continue. - Use the project's existing framework and conventions when supplied. Do not invent APIs, package versions, environment variables, or infrastructure that were not provided; state assumptions. - Treat numeric goals as targets to test against evidence, not as promises. REQUIRED OUTPUT 1. Assumptions 2. Design/approach 3. Implementation/code 4. Tests and edge cases 5. Run/verification steps Where alternatives are useful, provide no more than three materially different options and explain the trade-off of each; do not create cosmetic variants that do not change the decision. FAILURE PREVENTION Quality-check the result against: code is runnable with stated dependencies; failure cases are handled; business logic is testable. Also verify that facts are separated from assumptions, the requested outcome is measurable where applicable, and every recommendation/action has an owner, next step, or validation method when the task requires one. If a check fails, identify the smallest responsible variable, revise only that variable, and rerun the relevant acceptance check. Do not rewrite the entire solution just to make it look different. FINAL STANDARD The result must be usable by a professional in the stated Python context on the first serious execution. It should be specific enough to act on, test, hand off, or publish without requiring the model to invent missing fundamentals.