hddlmZmZmZddlmZddlmZmZm Z m Z m Z ddl m Z mZmZGdde ZGdde ZGd d e ZGd d Zy ))DictListUnion)Field) AtomicAgent AgentConfig BaseIOSchemaBaseToolBaseToolConfig) ChatHistorySystemPromptGeneratorBaseDynamicContextProviderceZdZUdZeed<eeeeeffed<eeefed<eeeeeffed<eeefed<eed<y) RecommendationInputSchemaax Schema for aggregated customer feedback for recommendations. Attributes: overall_sentiment_score: An integer score representing the overall sentiment. topic_sentiment: A dictionary mapping business topics to sentiment (e.g., {"Waiting Time": "negative"}). emotion_count: A dictionary counting emotions (e.g., {"joy": 2, "anger": 12}). words_sentiment: A dictionary mapping key words to sentiment (e.g., {"delicious": "positive"}). sentiment_breakdown: A dictionary representing the percentage breakdown of sentiments (e.g., {"positive": 70, "negative": 20, "mixed": 2, "neutral": 8}). overall_sentiment_scoretopic_sentiment emotion_countwords_sentimentsentiment_breakdown total_reviewsN)__name__ __module__ __qualname____doc__int__annotations__rstr!#tCH~-..c3h'rrceZdZUdZeddZeed<eddZeed<eddZ eed <edd Z e ed <y ) RecommendationRowScemazt_summary_ Args: BaseIOSchema (_type_): _description_ Returns: _type_: _description_ .z$Recommendation for a specific topic. descriptionrecommendationz+Topic for which the recommendation is made.topiczSeverity of the recommendation.severityzRank of the recommendation.rankN) rrrrrr%rrr&r'r(rrrr r"r"sQ 1WXNCXs(UVE3V#+LMHcMc'DED#Err"c6eZdZUdZeddZeeed<y)RecommendationOutputSchemaz Schema for business recommendations based on feedback. Attributes: recommendations: A dictionary mapping business topics to recommended actions. .zADictionary mapping business topics to actionable recommendations.r#recommendationsN) rrrrrr+rr"rrrr r*r*.s& 5: \5OT01rr*c.eZdZddedefdZdedefdZy) RecommendationAgentmodel temperaturectdggdgd}t||||}tttf||_y)a Initialize the RecommendationAgent. Args: client: The LLM client (e.g., from instructor.from_openai). model: The identifier for the model to use. temperature: Sampling temperature for the LLM. zQYou are an expert business consultant specializing in customer feedback analysis.)z>Review the aggregated customer feedback provided in the input.ztConsider the overall sentiment score, topic sentiment, emotion counts, words sentiment, and the sentiment breakdown.zvIdentify the areas needing improvement (e.g., if 'Waiting Time' is negative or if emotion counts indicate high anger).zlGenerate clear and actionable recommendations for each relevant topic or words to help the business improve.)zqReturn the output as a JSON object with a single key 'recommendations' which is a list of recommendation entries.zlEach entry in the list should be a dictionary containing: 'recommendation', 'topic', 'severity', and 'rank'.zInclude recommendations for each topic or word that shows significant negative or mixed sentiment, or important business opportunity.zKRank recommendations based on urgency and severity, where 1 is most urgent.uQDo not group multiple topics into one recommendation — each should be separate.) backgroundstepsoutput_instructions)clientr.r/system_prompt_generator)configN)r rrrr*agent)selfr4r.r/r5r6s r __init__zRecommendationAgent.__init__;sR#8kl ! # &#$;   !!:;U!UV^de raggregated_feedbackreturncRtdi|}|jj|}|S)a` Generate business recommendations based on aggregated feedback. Args: aggregated_feedback: A dictionary containing overall_sentiment_score, topic_sentiment, emotion_count, words_sentiment, and sentiment_breakdown. Returns: An instance of RecommendationOutputSchema with the recommendations. r)rr7run)r8r: input_dataresults r generate_recommendationsz,RecommendationAgent.generate_recommendationsbs)/E1DE  + rN)z gpt-4o-minig?) rrrrfloatr9dictr*r@rrr r-r-:s-%fc%f%fN D E_ rr-N)typingrrrpydanticr atomic_agentsrrr r r atomic_agents.contextr r rrr"r*r-rrr rGsM$$[Z``  & F\ F  44r