generator-begcode/dist/agent/functions/AnalyzeData.js

Spring Boot + Angular/React/Vue in one handy generator

import { ChatMessageBuilder, CsvChunker, Rag, ArrayRecombiner, Prompt, } from '../agent-core/index.js';
import { LlmAgentFunctionBase } from './utils/index.js';
export class AnalyzeDataFunction extends LlmAgentFunctionBase {
    constructor(llm, tokenizer) {
        super(llm, tokenizer);
    }
    name = 'analyzeData';
    description = 'Analyzes large datasets and returns a comprehensive summary of all details';
    parameters = {
        type: 'object',
        properties: {
            data: {
                type: 'string',
                description: 'The datasets to be analyzed',
            },
            question: {
                type: 'string',
                description: 'the question your analysis is trying to answer',
            },
        },
        required: ['data', 'question'],
        additionalProperties: false,
    };
    async analyze(params, context) {
        const chunks = CsvChunker.newlinesWithHeader(params.data, { chunkLength: 20, overlap: 2 });
        const relevantChunks = await Rag.standard(context)
            .addItems(chunks)
            .query(params.question)
            .recombine(ArrayRecombiner.standard({
            limit: 3,
            sort: 'index',
        }));
        let prompt = new Prompt();
        for (let i = 0; i < relevantChunks.length; ++i) {
            const chunk = relevantChunks[i];
            prompt = prompt.line(`Chunk #${i}`).block(chunk);
        }
        prompt = prompt
            .line(`
        Consider the above chunks of CSV data.
        Detail any column names & data formats used.
        Summarize the semantic meaning of the chunks.
        Tailor your response to the following question:
      `)
            .line(params.question)
            .line('BE VERY TERSE IN YOUR RESPONSE.');
        return this.askLlm(prompt.toString(), {
            model: 'gpt-3.5-turbo-16k',
            maxResponseTokens: 100,
        });
    }
    buildExecutor({ context, }) {
        return async (params, rawParams) => {
            const summary = await this.analyze(params, context);
            return {
                outputs: [],
                messages: [ChatMessageBuilder.functionCall(this.name, rawParams), ChatMessageBuilder.functionCallResult(this.name, summary)],
            };
        };
    }
}