Exclusive Interview with Nishchay Shah, Chief Technology Officer, CACTUS

Cactus Communications is at the forefront of making scientific research accessible worldwide through language services, communication, publishing services, and AI-powered products for researchers. CACTUS Labs, the R&D division, operates at the forefront of AI, NLP, and big data technology. It is a globally distributed company with a remote-first policy. Lately, the company has been creating solutions that are becoming one of the finest, most rapid interpreters and enhancers of scientific information and literature. Analytics Insight has engaged in an exclusive interview with Nishchay Shah, Chief Technology Officer, CACTUS.

Kindly brief us about the company, its specialization, and the services that your company offers.

Cactus Communications is at the forefront of making scientific research accessible worldwide through language services, communication, publishing services, and AI-powered products for researchers. We work with researchers, scientists, Nobel laureates, academic societies, life science companies, scholarly publishers, government bodies and think tanks from every part of the world. We build products powered by some of the most sophisticated AI-based language technology that exists.

CACTUS Labs, the R&D division, operates at the forefront of AI, NLP and bigdata technology. We are a globally distributed company with a remote-first policy. Lately, we have been creating solutions that are becoming one of the finest, most rapid interpreters and enhancers of scientific information and literature. We combine design, tech, and the best language experts to tackle real-world science problems. The products we create are helping great science come to the forefront in more significant ways than ever before.

Kindly brief us about your role at CACTUS and your journey

I oversee technology and innovation across products and brands globally at CACTUS. I am experienced in handling tech-budgeting, outsourcing, and global tech recruitment, and manage a large department with over 300 experts working in product management, software development, UX, DevOps, Digital innovation, and Machine Learning. I focus on creating, translating, and mobilizing big-picture visions downstream. I have over 17 years of experience in software development and technology and strive to stay on the bleeding edge of innovation. I've worked in the US for over a decade and handled a diverse team in the US, Belarus, Bulgaria, India, and the UK. Having successfully led both B2C and B2B product teams in the past, I have a thorough understanding of the end-to-end product and technology lifecycles.

Could you highlight your company's recent innovations in the AI/ML/Analytics space?

Academic Language Editing

Helps editors and researchers by taking care of grammar error detection + correction and providing sentence formation suggestions and potential reference checks from a corpus of millions of research artifacts

Academic Data Intelligence

We have deduplicated and cleaned nearly 300M+ academic articles and their metadata (Eg – journals, papers, clinical reports, etc) and created a large data lake

One of the most successful outcomes out of this data lake is R Discovery, our flagship product, which provides research paper recommendations for reading based on user profiles and macro + micro subject areas

Concept Extraction

The Concept Extraction engine facilitates the creation of content fingerprints and intent fingerprints, making it possible to group and match similar fingerprints with each other. It makes millions of documents accessible by leading users to where they want to be, going beyond brute force free text search.  R discovery, one of our apps powered by this engine, recently hit 1M+ downloads

Language and Phrase Bank Query Engine

It supports users during the drafting stage by helping them reference contextual word usage and grammar from published sources and make informed decisions

SoTA Language Model for Text Infiling for Academic Data Corpus

These are custom-trained large transformer models which are also in the range of billions of parameters. These are specifically trained on academic data and work much, much better than general-purpose language editors

Would you like to give some details about how Big Data Analytics/ML/AI/IoT is being used at CACTUS? How has it been progressing and benefiting the clients?

• Years' worth of editing data, collaborations with publishers, and setting up a massive data lake have helped us derive many AI/ML driven solutions
• All of the above solutions in the previous question are developed in-house at CACTUS which are based on AI, ML, and Big data
• Since CACTUS Labs was set up, we have grown each year with new products and solutions. This has helped create many of our current flagship products like R Discovery and Paperpal
• Internally as well, these solutions have helped different business units of CACTUS to unlock insights to break into new markets and make that gradual shift towards a product-driven hypergrowth
• Unclean, un-maintained data is one of the biggest problems we have seen, especially while dealing with academic data
• De-duplication, in-efficient data parsing, and complex ETL pipelines make it difficult to get results quickly
• One of the largest value adds we are bringing in is disambiguation, wherein we are working towards disambiguating all key metadata elements present in published articles and metadata

Can you throw light on the challenges faced in Big Data and Analytics industry?

AI as expert systems are evolving more towards assistive and pseudo-automatic stages in very shallow applications, given that people have realised it's better to solve small than to build generic systems. Recent trends emerging in the field are

Augmented Reality (AR) and smart glasses

Since the release of, among others, the Google Glass and Microsoft HoloLens applications, in the last few years, there have been significant advances in AR. This year, various organizations announced the release of their AR glasses products. Even companies like Ray-Ban, which are fashion-first, are foraying into smart glasses. These smart glasses allow people to interact and work in a real-world simulated environment. Over the next 5 years, we'll see.

Responsible And Ethical AI

If a self-driving car is faced with two choices, both of which result in some harm to a human, which decision should the model make? Should it be based on data OR should there be some override rule?  If a very novel advancement in AI has been made, is it okay for it to be used in a military application that will eventually be used in warfare?  These are some of the questions, along with bias, data protection, discrimination, etc., that responsible and ethical AI attempts to address. There is a strong movement around the ethical use of AI, and many companies are creating dedicated task forces and coalitions that deal with this.

AI Explainability

AI models, especially those that deal with larger derived dimensions of data and data gathered from various touchpoints, are largely deep-learning model black boxes. The data go in and the decision (output) comes out. There is very little reasoning behind why a certain decision was made. As we move into the future where AI is being used in applications such as medical diagnosis, self-driving vehicles, automated trading, and even in recruitment and other decision-making functions, it becomes important to ensure transparency and visibility on why a certain machine-learned model reached a particular decision. There are many open-source tools and frameworks that have yielded good early results in the interpretation of AI models.

• In critical applications like healthcare or disbursement of loans, we have seen how a lack of explanation can cause severe problems if human lives are impacted directly.

Fairness

• There have been many instances in different domains regarding AI systems being biased towards a certain group of people based on gender/ethnicity and other personal traits.
• These stem from a lot of actual data biases in the human world, reflected in data points, and many organisations are now starting to clean their datasets of these biases.

Privacy/ Federated Learning

• With regard to device information storage, the leaking of sensitive user data has marred quite a lot of AI applications, especially in the space of voice agents, facial recognition systems, and text-based conversations.
• Access to compute and the addition of more layers of engineering constraints makes it very difficult to deploy AI systems now if they are dealing with sensitive information.
• A lot of work is now picking up in this domain, with custom hardware created for faster AI compute gaining popularity, for example Mac M1, Google's TPU on a phone, and so on to support device calculations.