Hiring & training of people for GEN AI projects with Vikas Kamra

The hiring and training of talent for GEN AI projects are pivotal steps in harnessing the potential of this cutting-edge technology, ensuring that organizations can leverage it effectively to drive innovation and competitive advantage. Building a skilled workforce in GEN AI is essential for staying at the forefront of the AI revolution. 

In this week’s CDO Insights we have with us Vikas Kamra who is the Co-Founder & CEO of Valiance. He specializes in collaborating with global clients to create business value using data science, machine learning, and cloud technologies. Valiance has a successful track record of delivering projects for several renowned brands across various industries, including many Fortune 500 companies. These projects encompass areas such as Demand Planning, Customer-360, Personalization, Credit Risk & Fraud, Artificial Intelligence, Machine Learning, Generative AI, IoT Analytics, Data Quality and more. In addition to his professional work, Vikas is passionate about building high-performing teams in data science, data engineering, and software engineering. 

In this talk, Vikas discusses innovative approaches to building and managing Gen AI teams, emphasizing upskilling and unique talent acquisition strategies. He addresses the surprising absence of data scientists in Gen AI projects and the significance of job designations. He also touches on challenges in nurturing data science teams and highlights the importance of continuous learning for Gen AI professionals. Additionally, he explores hiring from non-Tier 1 colleges, initiatives for upskilling, and provides advice for data scientists interested in transitioning to Gen AI engineering.

AIM: In the realm of AI and Gen AI, there have always been AI teams, but with the advent of Gen AI and your involvement, what distinctive approaches or practices are you implementing to build and manage your team? How do you differentiate your talent acquisition and development strategies in this context?

Vikas Kamra: Well, it’s a great question. First of all, internally, we’ve had various skilled individuals working on AI projects. We have data scientists, MLOps engineers, ML engineers, and more. While there’s some overlap, everyone brings their unique expertise. For instance, ML engineers focus on taking ML algorithms and building continuous pipelines for algorithm deployment. Data scientists, on the other hand, primarily build and train models.

With the advent of Gen AI, which began gaining prominence with GPT-3 around November last year, and having reached 100 million downloads, it’s safe to say that practically everyone in the company, across all business functions, has had a taste of it. It seems like nearly everyone is utilizing such technologies in one way or another. Our business teams, for instance, use it for various purposes like creating email content and editing proposals. We’ve witnessed significant boosts in productivity.

When it comes to working on these projects, particularly with Gen AI, it’s important to note that the technology itself is relatively new, and there are continual developments. New frameworks emerge frequently, and we started with Lang chain for building document search and later moved to Llama index, which proved to be promising. So, fundamentally, we’re seeking strong software engineers, capable of rapid learning and adaptation.

With Gen AI, there are a few key aspects to learn. Understanding the technology itself is crucial. Prompt engineering, for example, is a vital skill that everyone needs to grasp. Whether you’re a product manager, a technology person, a salesperson or C-suite you need to comprehend how to work with this technology and use it effectively to get the job done. 

We provide in-house training to our team on the fundamentals of generative AI and it’s application for various use cases. This training is delivered through a combination of publicly available courses and hands-on exercises through internal projects. We conduct sessions not only for the tech team but also for product managers and delivery managers because it’s crucial for them to be well-versed with the technology, its capabilities, and its limitations. We delve into challenges such as implementing control over hallucinations and ensuring that answers are grounded in factual information. Additionally, we’ve established partnerships with hyper-scalers like Google Cloud and data platforms like Databricks. These partnerships have been instrumental in scaling our internal teams. It’s an exciting journey, and we’re committed to keeping our teams up to speed with the latest advancements.

AIM: Are you upskilling your current workforce for Gen AI or hiring externally? If hiring externally, what qualifications do you seek? If upskilling internally, which roles within your organization are you targeting for this transition?

Vikas Kamra: We did both – we hired externally and trained internally. When we hired externally, we looked for people with good aptitude and strong programming skills, particularly in software engineering and data structures. These were our primary criteria. Mostly, we hired fresh or recent graduates and provided them on the job training. As we continued to hire new talent, we found that many people were applying, making it challenging to filter resumes. So, we looked for individuals who had completed side projects involving Generative AI or had some internship experience in the field. While these were preferred qualifications, they were not strict filtering criteria. We emphasized strong aptitude and programming skills. This was our approach to external hiring.

Internally, we focused on training our software engineers. We provided them with courses on Generative AI, from Deeplearning.ai, Coursera & Google Cloud. These software engineers are well-versed in building applications and DevOps competency. Interestingly, we didn’t heavily involve data science skills in these projects; instead, we leaned toward software engineers and honed their skills.

This approach extended to our hiring for product managers. We assessed their proficiency in working with co-pilot technologies, such as ChatGPT. We have now incorporated Generative AI exposure into our hiring process for various roles. 

AIM: One thing that stood out is your mention that surprisingly you’re not hiring data scientists for Gen AI projects. In our research, we found that professionals working on Gen AI often don’t have specific job titles like ‘Gen AI Developer.’ How should companies effectively address this situation? Do job designations truly matter in this context, and if not, why?

Vikas Kamra: Designations do matter to people you see. ‘Data Scientist’ as a designation now is a lot clearer and mature. The next evolution I noticed was the ‘ML Engineer’ role responsible for the deployment and management of ML algorithms. ‘ML Engineer’ too, as a title has been in the market for a few years now. People understand what it entails. The next title I’ve come across is ‘AI Engineer.’ I find that this title is somewhat similar to the ‘ML Engineer,’ and when I look at resumes, I don’t see much differentiation. Recently, I noticed something on LinkedIn – when you post a job, there’s a specialist role in the dropdown menu called ‘Generative AI Engineer.’ Apart from ‘AI Engineer’ or ‘ML Engineer,’ which are well-recognized, you now also have this option in the dropdown. It seems like this title is also becoming mainstream, with many people using it to search for jobs. That’s likely why LinkedIn added it to the dropdown menu. In our team, we currently bracket them as ‘AI Engineers’ only. This allows us to differentiate them from traditional data scientists. However, Gen AI job titles are in their early stages, and if the technology changes, what title will we give to those individuals? Gen AI is relatively new, and we’re going through that hype cycle, building new things. So, assigning the title of ‘Generative AI Engineer’ brackets one person to one technology stream. We don’t prefer a scenario where a person is strictly limited to working on a particular type of AI technology, whether imposed by us or self-imposed. Gen AI doesn’t work in isolation; you can’t just write a prompt on Chat GPT and get things done. You have to write API codes and integrate them with third-party applications. In the end, a significant portion of the work is actually about putting the application together. Maybe around 70 percent of the work involves that. So, calling someone a ‘Generative AI Engineer’ or ‘Gen AI scientist’ may not be the right approach at this stage Present set of use cases and applications we are seeing mostly depend on consuming & fine-tuning third-party LLM’s hosted on the cloud. Hence now we will persist with the AI Engineer title and see how the market & technology evolves

AIM: In terms of talent acquisition and talent development, what specific challenges are you encountering in building and nurturing data science and analytics teams? Beyond hiring, can you provide more insights into the specific hurdles you’re facing and any strategies or approaches you’ve found effective in retaining and upskilling these teams?

Vikas Kamra: One particular hiring challenge we’ve encountered frequently, in virtual interviews conducted platforms like Google Meet, is candidates using ChatGPT to come up with answers. This can create the impression that a candidate is exceptionally skilled, potentially leading to hiring decisions based on misleading information. To address this, we’ve refined our interview process. Now, we ask candidates to share their entire screens, not just specific windows, particularly when coding assignments are involved. This adjustment helps us identify unusual behaviors, such as candidates opening mobile apps or streaming audio, and using tools like ChatGPT to provide answers. Additionally, we’ve reintroduced in-person interviews for certain cases, even if it means flying candidates to our office at our own expense, to ensure a more reliable evaluation.

Regarding skill development, the core challenges remain consistent across various technologies, including data science, ML engineering, and Gen AI. There’s a persistent shortage of qualified professionals in these domains, prompting companies to make substantial investments in upskilling and reskilling. Many organizations are dedicating significant resources and partnering with companies like Nvidia and Google Cloud to train and scale a large number of individuals, including those interested in Gen AI. While these efforts are expected to create a talent pool of entry-level professionals, true expertise in Gen AI is best acquired through hands-on experience.

AIM: Does hiring from non-Tier 1 colleges hold significance when it comes to Gen AI, given that cutting-edge technologies like Gen AI often involve research and proof-of-concept work, and how does this differ from the importance placed on Tier 1 institutions for traditional data science roles?

Vikas Kamra: I can explain why it makes sense for us. In our case, as a smaller company, we receive numerous applications from individuals interested in Gen AI positions. It becomes challenging to review every application thoroughly. Personally interviewing hundreds of applicants isn’t feasible. By focusing on Tier 1 colleges, we can narrow down the pool to a more manageable size, typically from 100 to about 30 applicants. This allows us to streamline the interview process, where we can assess aptitude and, subsequently, technical skills and cultural fit.

Additionally, we’ve found that talent from Tier 1 colleges often adapts quickly to our work culture and the fast-paced environment of Gen AI. While talent can certainly be found in Tier 2 or Tier 3 colleges, it may require interviewing a larger number of candidates to identify the right fit. We’re open to alternative approaches if there are more effective methods within the industry. The key is finding the right talent while maintaining the speed and efficiency required in our field, where we frequently experiment and deliver results within tight timeframes.

AIM: Continuing from that, these professionals are constantly engaged in learning, both on the job and through self-study. What initiatives is your company taking to facilitate their upskilling? Additionally, could you highlight some of the trending skills that you anticipate Gen AI professionals will need to acquire in the near future?

Vikas Kamra: I’ve noticed that with prompt engineering for one to two months, you might even say that you have a reasonable experience. But every next day, you will be thrown 10 challenges, and you will see the technology behaving in a weird way. We’re building some pilots and proof of concept. We create a prompt today, and the next day, we test it again with certain new inputs, and the output is completely weird. It can be such a black box that you don’t know why it’s behaving in such a manner. Even a slight change in the prompt can alter the output significantly. This makes you question whether the prompt you used will work reliably today or not. And even if it is reliable today, let’s say you’ve used the prompt with OpenAI’s GPT, is it portable to another model like PaLM? I don’t think so. Today, prompts aren’t portable across different models. The results vary, and we’ve tested the same problem with different models, yielding different results, which is expected. So, we have to fine-tune our prompts differently for these models. Prompt engineering thus in my view is one the most important skills required to get the best out of generative AI for any role. Apart from this there are foundational models, frameworks and technologies which help you create applications like Lang chain, Llama index, Vector databases etc.

Our training process is a combination of online courses and on-the-job training. Now that we have gained reasonable experience working on multiple projects, we do in-house workshops for new hires. Each team member is provided with access to the cloud environment where they can experiment with LLM’s and create GEN AI applications. New hires are also asked to build novel GEN AI applications under the guidance of a mentor.

To further help with upskilling, we give our team access to courses from Google Cloud and other data platforms. Once people are settled, we also encourage them to spend 8 to 10 hours per week in just learning new technologies & frameworks that can improve the solution offerings we are creating. This is further complemented by giving our team complimentary access to technology blogs, magazines, and books.

AIM: Do you have any key advice for data scientists who are interested in transitioning to Gen AI engineering and want to work on projects in that field?

Vikas Kamra: Our GEN AI team today primarily consists of software engineers, not data scientists. So, surprisingly, I haven’t witnessed a clear path where data scientists transition to become Gen AI engineers. While data scientists can certainly upskill themselves, especially through courses on Gen AI and deep learning, and develop key skills like prompt engineering, when it comes to building Gen AI applications, the focus is less on training models from the ground up and involves less heavy mathematics and architectural work. We primarily integrate third-party algorithms and technologies, with limited use of traditional data science skills. We’re not heavily involved in training models or complex data science tasks.

In terms of advice, data scientists can certainly enhance their capabilities by learning to use Gen AI as a co-pilot. For example, they can streamline data science tasks like Exploratory Data Analysis (EDA) by leveraging Gen AI’s capabilities to generate code and insights more efficiently. This allows them to become more effective and efficient in their current roles without necessarily transitioning to Gen AI engineering.

I believe that the demand for data scientists will persist because there will always be a need to interpret data in a meaningful and contextual way for businesses. While technology can assist with certain tasks, it may not fully grasp the intricacies of a specific business context. Therefore, data scientists can evolve by learning how to work effectively with Gen AI technologies to enhance their outcomes. This shift isn’t necessarily from data science to Gen AI engineering but rather an evolution within the data science domain, where data scientists leverage Gen AI as a valuable tool to improve their results. For instance, Gen AI can expedite tasks such as explaining clusters in data analysis, making processes more efficient.