You are currently viewing Correcting code: CodeMate's AI programming assistant debugs, fixes errors automatically

Correcting code: CodeMate's AI programming assistant debugs, fixes errors automatically


When it comes to writing an email or an essay, digital writing assistants make it easier to create a draft that is readable and without errors. One of these tools is Grammarly which helps users enhance the quality and accuracy of written content in English using artificial intelligence (AI) that offers suggestions on how to improve grammar and sentences.

Ayush Singhal wanted to build something similar for software code—a tool that would correct and upgrade the programming language as it is being coded by the developer. 

“I have been a developer and have previously owned a software agency. I had the first-hand experience of project timelines getting delayed due to bad code,” he tells YourStory.

In 2022, he founded CodeMate. The startup is a part of Tech30 list of most promising companies announced by YourStory at TechSparks 2023 in Bengaluru.

“CodeMate is an auto-correcting developer tool that fixes coding errors as you type,” says the Founder of the Noida-based startup. 

Grammarly for coding

CodeMate’s six-member team created a plugin which can debug code, including fixing syntax errors and run-time errors, as well as warning about performance issues, and more. The coding assistant analyses the code to determine the errors and suggest solutions. 

Singhal says that CodeMate can review code like an experienced developer. 

It can inform if there is sensitive information in the program such as API keys, which can be used by bad actors to gain access to servers. The tool can also suggest removing coding redundancies once testing is over to make it cleaner—considered good practice in the industry

Users can also limit the review to predefined parameters to focus on certain sections of the code. It also gives the code a score out of 100 on parameters including time and space complexity and benchmarked data to determine if the code is accurate and clean.

Programmers can also add knowledge bases and use the chatbot functionality to obtain information from them to solve certain enquiries. In the context of coding and software development, a ‘knowledge base’ refers to a centralised repository or database that stores information, documentation, and resources related to a programming language, framework, library, or any other technical aspect.

The chatbot can tell users how to include a certain API or determine defined paths and their purposes. If CodeMate determines the knowledge base has the appropriate code for the task, it also nudges the user to import the code and integrate it into the program.

“Through context-aware suggestions and fixes, and the ability to train on your repositories, CodeMate is like an experienced developer sitting beside you and having all the knowledge about your coding style, current codebase and problem description,” Singhal says, adding that programmers can solely focus on building products, leaving the non-productive tasks like debugging to AI.

<figure class="image embed" contenteditable="false" data-id="537082" data-url="https://images.yourstory.com/cs/2/464cfc80fc4b11ec863f3b2095783040/Imagewplp-1684075907852.jpg" data-alt="CodeMate founder" data-caption="

Ayush Singhal, Founder of CodeMate, in an earlier conversation with YourStory Founder and CEO Shradha Sharma

” align=”center”>CodeMate founder

Ayush Singhal, Founder of CodeMate, in an earlier conversation with YourStory Founder and CEO Shradha Sharma

@media (max-width: 769px)
.thumbnailWrapper
width:6.62rem !important;

.alsoReadTitleImage
min-width: 81px !important;
min-height: 81px !important;

.alsoReadMainTitleText
font-size: 14px !important;
line-height: 20px !important;

.alsoReadHeadText
font-size: 24px !important;
line-height: 20px !important;

Also Read

With AI-based cloud observability, Middleware helps developers move over debugging

All about numbers

Revenue in the software market is projected to reach $698.80 billion in 2024, according to Statista. Singhal estimates that the software industry burns $300 billion every year in just fixing maintenance issues, and developers have to spend upward of 1,500+ hours annually debugging code.

“Through CodeMate, organisations can save up to 40% on their project cost and 35-50% in their project time,” he claims.

CodeMate operates on a subscription-based revenue model, offering three distinct plans. It targets corporates, individual contributors, and freelancers.

The Individual subscription follows a freemium model, providing basic features for free and premium features starting at Rs 1,200/month. CodeMate for Education caters to educational institutions, offering tailored features for evaluation and programming scores. The Enterprise subscription targets businesses and development teams, providing code review and debugging features.

Currently, the startup has onboarded 15 companies as well as 45,000 individual users. The company is also planning a couple of pilots with universities in the coming months. In the last three months, the company has earned over $50,000 from its clients.

Standing apart

However, rapid developments in large language models (LLMs) may spell serious trouble for CodeMate. Companies have already started making entire apps and websites using just OpenAI’s ChatGPT. GitHub Copilot is another AI pair programmer that offers coding suggestions and functions while typing, powered by a massive dataset of public code.

However, Singhal says that CodeMates’s context awareness makes it stand apart. 

Companies often have separate code bases for their web and mobile applications. Different teams may be responsible for different codebases, allowing for more efficient development and maintenance. However, managing multiple codebases can also be challenging, especially when it comes to ensuring consistency and avoiding duplication of effort.

“A metaphor we have used is that a raw LLM is like a booksmart programmer who has read all the manuals but does not know about a companyʼs codebase. By providing context from a company codebase along with the LLM prompt, the LLM can generate an answer that is relevant to that codebase,” he notes.

CodeMate can refer to multiple codebases and provide suggestions to make sure the code is maintainable, scalable, and adaptable. In contrast, GitHub Copilot’s context length is only 2048 tokens, or 150-200 lines of code.

Singhal emphasises that while ChatGPT handles code debugging, he asserts that tools wrapped around it exhibit a substantial amount of “hallucinations.” He clarifies that the team is developing an autonomous AI agent within CodeMate.ai. This agent not only generates fixed code but also autonomously corrects its own code iteratively until it successfully passes all test cases on the backend. The tools offer suggestions only after this rigorous process.

“We will continually enhance our AI-powered coding assistance with new features, language support, and framework compatibility. This will attract a wider developer audience and solidify our position as an essential coding tool,” the founder adds.

The company, which is bootstrapped with an undisclosed amount, is also developing a voice-enabled feature that allows users to interact with the code solely using their voice. This feature is scheduled for launch next month, initially catering to English speakers. It plans to extend the same functionality to Indian vernacular languages in the coming months.


Edited by Kanishk Singh



Source link

Leave a Reply