mongodb-rag-core/build/mongoDbMetadata/classifyMetadata.js

Common elements used by MongoDB Chatbot Framework components.

"use strict";
Object.defineProperty(exports, "__esModule", { value: true });
exports.tagifyMetadata = exports.classifyMongoDbMetadata = exports.classifyMongoDbProgrammingLanguageAndProduct = exports.classifyMongoDbTopic = exports.ClassifyMongoDbTopicSchema = exports.classifyMongoDbProduct = exports.ClassifyMongoDbProductSchema = exports.classifyMongoDbProgrammingLanguage = exports.ClassifyMongoDbProgrammingLanguageSchema = void 0;
const zod_1 = require("zod");
const aiSdk_1 = require("mongodb-rag-core/aiSdk");
const products_1 = require("./products");
const programmingLanguages_1 = require("./programmingLanguages");
const topics_1 = require("./topics");
const braintrust_1 = require("braintrust");
const baseSystemPrompt = `You are an expert data labeler employed by MongoDB.
You must label metadata based on the provided label categories:`;
exports.ClassifyMongoDbProgrammingLanguageSchema = zod_1.z.object({
    programmingLanguage: zod_1.z
        .enum(programmingLanguages_1.mongoDbProgrammingLanguageIds)
        .nullable()
        .describe("Most important programming language present in the content. If no programming language is present, set to `null`."),
});
/**
  Classify metadata relevant to the MongoDB docs chatbot
  from a user message in the conversation.
 */
exports.classifyMongoDbProgrammingLanguage = (0, braintrust_1.wrapTraced)(async (model, data, maxRetries) => (await (0, aiSdk_1.generateObject)({
    maxRetries,
    messages: [
        {
            role: "system",
            content: `${baseSystemPrompt}

${programmingLanguages_1.mongoDbProgrammingLanguages.map((l) => ` - ${l.id}: ${l.name}`).join("\n")}`,
        },
        {
            role: "user",
            content: data,
        },
    ],
    model,
    schema: exports.ClassifyMongoDbProgrammingLanguageSchema,
})).object.programmingLanguage, {
    name: "classifyMongoDbProgrammingLanguage",
});
exports.ClassifyMongoDbProductSchema = zod_1.z.object({
    mongoDbProduct: zod_1.z
        .enum(products_1.mongoDbProducts.map((p) => p.id))
        .nullable()
        .describe(`Most important MongoDB product present in the content.
If it is not clear what the product is, set to \`null\`.`),
});
exports.classifyMongoDbProduct = (0, braintrust_1.wrapTraced)(async (model, data, maxRetries) => (await (0, aiSdk_1.generateObject)({
    maxRetries,
    messages: [
        {
            role: "system",
            content: `${baseSystemPrompt}

${products_1.mongoDbProducts
                .map((p) => ` - ${p.id}: **${p.name}**. ${p.description}`)
                .join("\n")}

Keep in mind:
1. Include "Driver" if the user is asking about a programming language with a MongoDB driver. 
2. If asking about MongoDB database but not specifying if Atlas/Server/Op Manager, etc, set to "MongoDB Server".
3. If not clear what product is being asked about, set to \`null\`.`,
        },
        {
            role: "user",
            content: data,
        },
    ],
    model,
    schema: exports.ClassifyMongoDbProductSchema,
})).object.mongoDbProduct, {
    name: "classifyMongoDbProduct",
});
exports.ClassifyMongoDbTopicSchema = zod_1.z.object({
    topic: zod_1.z
        .enum(topics_1.mongoDbTopics.map((t) => t.id))
        .nullable()
        .describe(`Most important MongoDB-related topic present in the content. If none of the listed topics are clearly present, set to \`null\`.`),
});
exports.classifyMongoDbTopic = (0, braintrust_1.wrapTraced)(async (model, data, maxRetries) => (await (0, aiSdk_1.generateObject)({
    maxRetries,
    messages: [
        {
            role: "system",
            content: `${baseSystemPrompt}

${topics_1.mongoDbTopics
                .map((t) => ` - ${t.id}: **${t.name}**. ${t.description}`)
                .join("\n")}`,
        },
        {
            role: "user",
            content: data,
        },
    ],
    model,
    schema: exports.ClassifyMongoDbTopicSchema,
})).object.topic, {
    name: "classifyMongoDbTopic",
});
function nullOnErr() {
    return null;
}
exports.classifyMongoDbProgrammingLanguageAndProduct = (0, braintrust_1.wrapTraced)(async (model, data, maxRetries) => {
    const [programmingLanguage, product] = await Promise.all([
        (0, exports.classifyMongoDbProgrammingLanguage)(model, data, maxRetries).catch(nullOnErr),
        (0, exports.classifyMongoDbProduct)(model, data, maxRetries).catch(nullOnErr),
    ]);
    return { programmingLanguage, product };
}, { name: "classifyMongoDbProgrammingLanguageAndProduct" });
exports.classifyMongoDbMetadata = (0, braintrust_1.wrapTraced)(async (model, data, maxRetries) => {
    const [programmingLanguage, product, topic] = await Promise.all([
        (0, exports.classifyMongoDbProgrammingLanguage)(model, data, maxRetries).catch(nullOnErr),
        (0, exports.classifyMongoDbProduct)(model, data, maxRetries).catch(nullOnErr),
        (0, exports.classifyMongoDbTopic)(model, data, maxRetries).catch(nullOnErr),
    ]);
    return {
        programmingLanguage,
        product,
        topic,
    };
}, {
    name: "classifyMongoDbMetadata",
});
// Type guard to check if a value is a non-null MongoDbTag
function isMongoDbTag(value) {
    return value !== null;
}
/**
  Convert metadata to tags
 */
function tagifyMetadata(metadata) {
    return Object.values(metadata).filter(isMongoDbTag);
}
exports.tagifyMetadata = tagifyMetadata;
//# sourceMappingURL=classifyMetadata.js.map