What if …? How about ...? Why not …?
At Quantrium Labs, every project starts with a question and culminates in deepening our AI expertise.
Quantrium Labs is a vital element of Quantrium’s research initiative. It strengthens our competence to tackle novel and difficult challenges — like document intelligence for an American healthcare analytics company. And near-100% accurate data extraction from paystubs for a leading Asian mortgage provider.
With lead engineers as mentors, the young team at Labs gets help with pointers to relevant research literature and granular definitions of problems.
Ideas incubated in Quantrium Labs have also facilitated the company’s entry into new markets. A video analytics project that started as a fun exercise in Labs has fueled a more serious exploration of AI applications for the purpose of ecological conservation.
The questions continue to pop up.
And the team at Labs loves to take a crack at them.
Here’s a peek at what’s cooking in Labs
Based on BigBird and BERT, the NLU engine under development is programmed to assume human-like text summarization capabilities. This will power a new generation of intelligent chatbots and email bots.
What if AI writes its own rules to analyze the data instead of following what’s pre-set by humans? The project explores revolutionary AI concepts to identify relationships among features and dynamically devise rules. The learnings will lead us to the next generation of AI.
This project is aimed at leveraging a blend of robust image recognition and NLP algorithms to distinguish individual documents in a file. The approach will enhance record storage and retrieval processes.
Powered by optimized text-extraction methods and a suite of NLP models, the project provides the capability to identify the structure of a document. This facilitates intelligent document digitization.
The AI Scraper is being developed to understand the changes in a web page and modify the scraping scripts on the fly so as to continue scraping the website content with minimal disruption.
With customized CNNs and auto-encoder, it is targeted to detect and identify the objects in a video at low frame rates. It will associate the detections with the predicted new locations in a new frame with Temporal context. This will bring a new dimension to Object tracking in videos at low frame rate while achieving high accuracy.