Overview

How does Meldus work?

Meldus’s powerful AI engine is purpose-built to handle the complexity of Salesforce data and to answer your questions as simply and naturally as if you were asking a colleague. Behind the scenes, however, Meldus performs multiple sophisticated processes to interpret, translate, and summarize your questions with precision and accuracy.

Here’s a closer look at how it all works.

Interpreting your question

Meldus begins by interpreting your question’s intent; parsing your natural language to recognize its nuances and implicit meaning. Meldus understands variations in phrasing and context, even when specific concepts aren’t directly stated. For example, Meldus understands that “recent monthly performance” implicitly references relative date literals and comparisons.

Classifying based on meaning and topic

Meldus classifies your question according to its semantic meaning and topical categories. Meldus's language models go well beyond keyword matching, to classify your question and recognize different modes of analysis and data domains. Our models distinguish, for example, between requests for statistical summaries, growth patterns, or breakdowns across data subsets, even when phrasing is ambiguous.

Breaking the question into components

Next, Meldus dissects your question into its key components, isolating each required data point. This step parses entities like dates, quantities, and conditions, so each part can be linked to the correct Salesforce data. Meldus is designed to handle complex questions involving multiple conditions, such as “How did revenue grow in Q1 compared to Q4 last year for the top five accounts?” which demands precise identification of multiple, interrelated data points.

Mapping to Salesforce objects and fields

Meldus then maps each component to the correct Salesforce objects and fields. Since each Salesforce instance can vary in structure, with custom fields alongside standard ones, this step requires real-time retrieval of metadata to understand how your Salesforce environment is configured. Meldus dynamically adapts to customizations, to understand, for example, if “Revenue” refers to a custom field, or is used as a synonym for opportunity amount.

Translating into Salesforce-compatible commands

With all components in place, Meldus translates your question into technical commands that Salesforce can execute. This translation involves complex logic to ensure that the query captures each nuance of the question components, such as aggregations, filters, or nested conditions. Additionally, Meldus ensures that all fields are formatted correctly and cross-references relationships across multiple objects, even when the question involves indirect or inferred relationships.

Retrieving precise results from Salesforce

Once the commands are constructed, Meldus executes them against Salesforce to retrieve real-time data. This process includes optimization techniques to handle datasets efficiently and to avoid timeouts or incomplete data, even when the query involves cross-object references, aggregations, or complex filters.

Summarizing into natural language

Finally, Meldus synthesizes the raw data from Salesforce into a clear and concise executive summary, using natural language generation to interpret and phrase the results conversationally. This step is technically challenging, as it requires Meldus to accurately interpret numerical data, spot patterns, and contextualize findings, ensuring that insights are accurate and useful, at the right level of fidelity. Meldus endeavors to highlight what's import in your results from Salesforce, making complex data easy to understand at a glance.

The end result?

You can ask questions naturally, and Meldus handles the intricate technical steps — from parsing language to querying complex data structures and generating polished summaries — so you can focus on insights rather than the technical details.