[Tech30] How Taskmonk is helping AI/ML enterprises optimise human efforts and control data labelling budget
Building successful AI models require massive amounts of labelled data. But data labelling is an intensive process that requires a lot of human effort. For enterprises, this effort is magnified many folds as they have multiple fronts to deal with: labelling projects, use cases, data types, and labelling teams. Overall, this leads to a significant increase in the data labelling budget.
To deal with such issues, Taskmonk optimises the conversion of raw data to labelled data, thus enabling enterprises to optimise - human effort, control – labelling budget, and enhance – the quality of labelled data.
The founding journey
Chetan Velkur (38), Sampath Herga (43), and Vikram Kedlaya (30) co-foundedin July 2018.
Chetan was running an impact sourcing BPO that was performing data labelling projects for large enterprises. He was facing problems in developing the right technology which would enable him to run different data-labelling projects on a single platform. Towards the beginning of 2018, Chetan reached out to Sampath who had built an AI startup (Meshlabs, which was acquired by Pega in 2014) and had seen the problem from the other side of the equation in generating training data for his AI models.
Together, they realised that providing a platform for data annotation will help AI/ML companies to execute and manage data annotation projects across annotation partners, and also empower them to annotate over a highly configurable best practice platform.
Vikram, who had worked with Sampath in his earlier startups to build end-to-end products, was roped in to lead the development efforts.
How does it work
Taskmonk’s Proprietary task allocation algorithms and ML assisted labelling can increase human labelling speed by upto 60 percent.
The typical flow is that an enterprise starts with large amounts of data that they need to be labelled. The labelling could be of various types – a piece of text or image that needs to be classified or entities identified; an image that needs to have objects annotated by drawing bounding boxes; a video that needs to have scenes classified or objects tracked across the video; or an audio file that needs to be transcribed, translated, transliterated, and categorised.
Enterprises use Taskmonk to connect with teams that would be doing the actual labelling work on the data. They upload the data to the platform and once the labelling teams finish the work, the results are returned to the enterprise. The enterprise would then use the labelled data to build their AI models.
Taskmonk has till date delivered 44 million tasks and enabled enterprises to save more than 325,000 human labelling hours.
Taskmonk runs on a software-as-a-service (SaaS)-based subscription model. It charges a per-hour usage fee for the platform. It claims to make a monthly recurring revenue of $25,000, or annual recurring revenue of $300,000.
Taskmonk targets BPO firms that do the labelling work and customers that want training data generated.
“This is extremely beneficial for domains like ecommerce that are developing AI models for use cases ranging from competitive intelligence, (Text/Classification Algorithms), visual search (Image/Computer Vision) and customer support bots (Audio/NLP),” adds Sampath, Co-founder and CEO.
Taskmonk’s competitors include companies such as Scale, LabelBox,, and , with most of them based out of North America. Speaking about what makes the startup stand out from its competitors, Chetan, Co-founder and CPO, says,
“Taskmonk supports more data types for labeling. Also, we are the only platform that enables enterprises to manage multiple labelling teams for the same project. Most of our customer say that the platform is built from the labeller perspective, thus enabling labelling teams to generate high-quality work.”
Currently, they have three big clients that include, IndiVillage, and , and plans to add three more customers in the next quarter.
Taskmonk, being a horizontal solution provider across the AI sector, primarily focused on platforms including automotive, ecommerce and retail. It has started expanding to other domains like Conversational AI. The startup is also planning to expand to East Asia and Europe.
Edited by Kanishk Singh