Hello everyone 👋

It’s been an exciting year for Catalyst QuickML.

In 2025, as the Catalyst platform continued to expand its capabilities, QuickML focused on strengthening reliability, developer control, and operational readiness across the AI lifecycle. This year marked an important step forward in making AI easier to adopt, scale, and operationalize within the Catalyst ecosystem.

For those who do not know:

QuickML is the AI/ML layer of the Catalyst platform, offering AI as a Service to help organizations of all sizes transform ideas into production-ready AI solutions at scale. By abstracting infrastructure complexity and technology overhead, QuickML enables faster experimentation and seamless deployment of scalable, intelligent applications through an intuitive no-code experience.

Do checkout the Catalyst QuickML here.

Built on the pillars of data security and customer trust, QuickML democratizes AI by empowering teams to design, build, and deploy end-to-end AI solutions with transparency and confidence. Over the past year, we expanded QuickML’s capabilities across Custom Model development, Generative AI features, and performance optimizations for faster model training and deployments, and enhanced the product experience with more intuitive interfaces.

Our focus remains on delivering a product that is simple to use yet powerful enough to support the full spectrum of machine learning requirements for our customers.

Here’s a look back at what we shipped in 2025, followed by a sneak peek of What’s coming next and Real-World Use Cases built on QuickML:

New Additions

1. Generative AI – A new era of intelligent workflows

This year, we launched a dedicated Generative AI module, opening up the ability to explore and operationalize advanced language models. It module is equipped with:

a. LLM Serving & Integration: 

QuickML now hosts a series of language models from Qwen, modified for multiple use cases. Now businesses can deploy and interact with these powerful models directly in QuickML’s chat interface. Available models are:

1. Owen 2.5 - 14B Instruct

2. Qwen 2.5 - 7B Coder

3. Qwen 2.5 - 7B Vision Language Model

Tune outputs with configurable parameters such as Temperature, Max Tokens, etc. depends on business requirements and integrate securely with OAuth-based authentication. | Help Document.

LLM Serving

b. RAG (Retrieval-Augmented Generation): 

Deliver context-aware natural language responses by combining user queries with real-time information from your organization’s knowledge base — ensuring more relevant and accurate results. | Help Document.

RAG

c. Knowledge Base: 

A centralized repository for uploading, organizing, and managing your organization’s key documents - acting as a single source of content database. | Help Document.

Knowledge Base

Why does it matter?

_{While Generative AI has unlocked powerful new possibilities, productionizing GenAI within business systems and applications remains complex and often comes with high costs and operational overhead. QuickML addresses this challenge by offering a dedicated Generative AI module that supports multiple LLMs, enhanced RAG capabilities, and a built-in knowledge base to serve as a centralized data repository.}

_{QuickML provides a secure and intuitive environment to test, build, and deploy the natural language layer using configurable LLM and RAG parameters tailored to specific requirements. Through QuickML’s intuitive chat interface, teams gain greater control and visibility, enabling them to move from experimentation to scalable, real-world applications faster and with confidence.}

2. Smarter insights with vision language model

We embedded chart insight generation using the Qwen 2.5 - 7B Vision Language Model, helping you automatically detect trends, patterns, and anomalies in your visualized data — no manual analysis required. | Help Document.

Chat Insights

Why does it matter?

Generating contextually accurate insights using a predefined prompt and an image as input to the VLM as implemented here demonstrates how required information can be extracted in natural language format from just visual data. It highlights how users can gain deeper understanding by analyzing images and generating key insights through QuickML’s Vision Language Model.

3. New machine learning operations

We have added new operations to help you build more accurate and reliable ML models:

• Robust Scaler: Traditional scaling methods can be sensitive to extreme values (outliers), which can skew results and decrease the model performance. The new Robust Scaler uses statistics that are less affected by outliers, ensuring that your features are scaled more consistently — leading to more stable and trustworthy models. | Help Document.

• Class Imbalance Handling: Imbalanced datasets can bias models toward the majority class, resulting in poor predictions for minority classes. Our new class imbalance techniques automatically reweight or resample your data to balance class distribution, improving recall, precision, and overall model fairness - especially useful for fraud detection, churn prediction, and rare event modeling. | Help Document.
  

4. Early access features 

These capabilities were rolled out this year in Early Access and are already live for users.

We have introduced two major machine learning models that unlock deeper insights from your data:

a. Clustering models 

Quickly build clustering pipelines to identify natural segments and hidden patterns within your datasets. Whether you are grouping customers based on behavior or segmenting products, the new clustering capabilities make it easy to discover meaningful insights without any labeled data.| Help Document

Below is the chart generated to show the distributions of clusters generated by the model.

Cluster Plot

b. Anomaly detection

Detect unusual activity with both time-series and non–time-series anomaly detection models. These tools help you automatically flag outliers, spot fraud patterns, detect operational issues, and uncover rare events — enabling more proactive and data-driven monitoring across your workflows. | Help Document.

Below is the time series plot highlighting the anomalies detected by the model and the range of value thresholds at each data point.

Anomaly Detection plot

c. Custom code operations

We have significantly enhanced the custom code node, giving developers far more flexibility and control over pipeline logic. The Custom Code operations in the QuickML pipeline allow developers to embed their own Python logic directly into the model training process. By implementing Python classes provided through QuickML’s built-in templates, users can fully customize how data is transformed, how features are engineered, and even define the machine learning algorithm itself.

This capability is structured into three distinct components, custom data transformation, custom ML transformation, and custom algorithm, each targeting a different stage of the ML workflow: | Help Document.

Below is the prediction pipeline with a dedicated list of custom code operations highlighted a custom algorithm node with its predefined code template:

Custom code operations

5. Platform enhancements

1. Enhanced Visualization with Spark: We have integrated Spark to boost performance during chart visualizations and data preprocessing. This allows efficient handling of large datasets in visualization workflows, especially in chart rendering and preprocessing tasks.

2. Schema-Aware Node Configurations: Control dependent nodes with schema-aware configuration management. When a node is modified, QuickML detects impacted successor nodes and prompts you to reset or reconfigure them before execution. This ensures transparency, minimizes disruptions, and enables smoother data flow in complex pipelines. | Help Document.

Below demonstration shows the attempt to change the Select/Drop stage configuration resulting in highlighting the configuration changes of subsequent affected stages with specific user actions:

3. Improved Cost Tracking: Integrated QuickML usage stats with Catalyst’s billing panel for better visibility and cost control.

4. Expanded Data Import:

• Zoho CRM: Now supports importing data from sub-forms for richer analysis.

• Zoho Creator: Increased data import limit to 1.5M records (from 200K).

Customer use cases addressed using QuickML

Converted PDFs into images, classified them, and processed the visuals using Vision Language Models (VLMs) to extract required information and  structured as JSON responses.

2.Multilingual document processing:

Identified the language of PDF documents and extracted specific information  using prompt-based instructions specific to each page.

3.Speech Intelligence:

Transcribed audio in its original language, translated it to English, and performed sentiment, intent, and emotion analysis on the translated text to derive actionable insights from customer.

4.Automated Data Extraction & updation:

Extracted information from emails, spreadsheets, and PDFs, structured it, performed custom field mapping, updation in customer databases as per their requirements.

5.Customer Lifetime Value insights:

Generated natural language explanation about the influencing factors that are affecting Customer Lifetime Value based on the business & user data and Provided estimated range of CLV based on individual user profiles.

6.Demand Forecasting and Price Optimization:

Addressed intermittent demand and zero-inflated sales patterns by applying different forecasting models at a per-product level, followed by optimal price point estimation to maximize revenue potential.

What’s Coming Next?

As we move into 2026, QuickML will evolve into a powerful environment. It is getting enhancements to support the developers who want full control, flexibility, and hands-on depth in end-to-end AI model development process. We are preparing to unlock a new wave of AI development capabilities and integrations, and we invite every builder to get ready for what’s coming.

With this renewed momentum, we are taking the next significant step toward making QuickML, an AI hub for businesses.

Here’s a glimpse of what’s on the horizon:

1. Vision AI - Multimedia support

QuickML will soon enable teams to build advanced vision models that support image classification, object detection, and document intelligence. By extracting structured information from images and documents, organizations will be able to automate recognition tasks, streamline document processing, and unlock entirely new categories of AI-driven workflows.

Vision AI support marks a major extension of QuickML’s capabilities and opens the door to more sophisticated real-world applications.

2. Enhanced data import flow

A completely upgraded data import experience will allow you to:

• Upload multiple datasets simultaneously.

• Perform advanced data transformations, preparations, and modifications before bringing data into QuickML.

• This will streamline preprocessing and reduce manual overhead before model training.

3. Ready-to-use models

We are introducing a new suite of foundational models, each supported by dedicated SDKs and APIs, enabling organizations to integrate advanced AI capabilities directly into their applications. Ready-to-use models from QuickML reduces time to value by making it easy to add an intelligence layer to the application and operationalize AI.

As of early 2026, foundational models across Image processing, Computer vision, Text and Audio models are going to be live allowing businesses to instantly access these universal AI capabilities without building them from the scratch — just plug them into your workflows and use.

4. LLM Endpoints

LLM endpoints in QuickML will soon enable organizations to deploy dedicated LLM instances optimized for multiple use cases, including reasoning, coding assistance, question answering, image analysis, and retrieval-augmented generation. Each instance is provisioned with configurable parameters and required prompts, exposed via secure endpoint URLs and OAuth mechanisms to make it easy to integrate Generative AI capabilities into their production systems.

These updates are only the beginning, with many more advanced features & improvements planned and are in progress.

Wrapping Up !

2025 was a transformative year for QuickML with major advancements in generative AI, ML operations, performance optimization, and data handling. As we look ahead, 2026 is shaping up to be even bigger with its advanced AI capabilities, Out-of-the-box foundational models, expanding the scope of Gen AI integrations, support for more language models and more powerful RAG capabilities unified within a single platform.

We would also like to mention that none of these advancements would have been possible without the dedication and hard work of the entire QuickML team, from the developers who engineered these capabilities, to the QA experts, designers, and product experts who ensured every feature met our standards. Their commitment has played a crucial role in making QuickML smarter and more reliable for all our users.

If you haven’t explored Catalyst QuickML yet, now’s the ideal time to explore. Try out the new features and let us know your thoughts. We would love to hear your feedback and suggestions in the comments.

Have questions or need assistance? Feel free to reach out to our support team. We are here to help.

Stay tuned for more updates!

Thanks, and have a great day!

 
#Catalyst #QuickML