Iterate releases AppCoder LLM: Natural Language Generation AI for Code

In the current era where almost every enterprise is exploring how to leverage AI to drive business growth, vendors are competing to launch new and profitable tools that enable customers to easily build high-performance AI and machine learning-based applications.

Currently, low-code development is the main focus, but Iterate is taking steps to completely eliminate the coding layer. The California-based company, known for building and deploying AI and emerging technologies for private, edge, or cloud environments, recently announced the launch of AppCoder LLM, a finely-tuned model that uses natural language prompts to instantly generate working and updatable code for production-grade AI applications.

AppCoder LLM is integrated into Iterate's Interplay application development platform, which, like any other generative AI collaboration tool, uses text prompts and performs better compared to existing AI-based coding solutions, including Wizardcoder. This allows development teams to quickly obtain accurate code for products such as object detection or document processing.

"This innovative model can generate functional code for projects, greatly accelerating the development cycle. We encourage development teams to explore Interplay-AppCoder LLM and experience the powerful functionality of using our model to automatically generate code," said Brian Sathianathan, CTO of Iterate.ai, in a statement.

What sets AppCoder LLM apart?

At its core, Iterate's Interplay is a fully containerized drag-and-drop platform that connects AI engines, enterprise data sources, and third-party service nodes to create a workflow for production-ready applications.

Development teams can customize code writing for each node on this interface, and this is where AppCoder comes in. It allows users to generate code simply by giving instructions in natural language.

"Interplay-AppCoder can handle computer vision libraries, such as YOLOv8 for building advanced object detection applications. We can also generate code for LangChain and Google libraries, which are among the most commonly used libraries for chatbots and other functionalities," said Sathianathan.

For example, a fast-food delivery restaurant can connect a video data source and simply ask Interplay-AppCoder to write a vehicle recognition application using the YOLOv8 model from the Ultralytics library. LLM will immediately generate the required application code.

Sathianathan pointed out that his team was able to build a core, production-ready detection application in less than five minutes when testing this functionality. This accelerated application development can save costs and increase team productivity, allowing them to focus on strategic initiatives critical to business growth.

AppCoder excels in code generation with its leading LLM functionality

In addition to its speed, AppCoder LLM outperforms Code Llama and Wizardcoder in terms of output compared to Meta's Code Llama and Wizardcoder.

Specifically, in an ICE benchmark test, running the AppCoder and Wizardcoder models with a 15 billion version using the same LangChain and YOLOv8 libraries, Iterate's model scored 300% higher in functional correctness (2.4/4.0 compared to 0.6/4.0) and 61% higher in practicality (2.9/4.0 compared to 1.8/4.0).

A higher functional correctness score indicates that the model performs better in unit testing when considering the given problem and reference code, while the practicality score indicates that the model's output is clear, presented in logical order, and maintains readability, covering all the functionalities stated in the problem statement after comparing with the reference code.

"When generating code using an A100 GPU, the response time of Interplay-AppCoder is typically 6-8 seconds. The training process follows a conversational question>answer>question>context approach," added Sathianathan.

He noted that these results were achieved by fine-tuning the CodeLlama-7B, 34B, and Wizard Coder-15B, 34B models based on an understanding of LangChain, YOLO V8, VertexAI, and many other modern generative AI libraries.

Future developments

While AppCoder is now available for testing and use, Iterate states that this is just the beginning of their efforts to simplify enterprise-level AI/ML application development.

The company is currently building 15 private LLMs for large enterprises and focusing on applying these models to CPU and edge deployments to drive scalability.

"Iterate will continue to provide platforms and an expanded toolkit for managing AI engines, emerging language models, and large datasets, all fine-tuned for rapid development and deployment of (applications) on CPU and edge architectures. New models and data stacks are emerging, and our low-code architecture can quickly adapt and integrate these emerging models. This field is rapidly expanding and democratizing, and we will continue to drive innovative management and configuration tools into the platform," said the CTO.

Over the past two years, Iterate's revenue has nearly doubled. The company's clients include Fortune 100 companies in sectors such as banking, insurance, document services, entertainment, luxury goods, automotive services, and retail.