At Matrix Partners India, we recently organised a session on GenAI for our 50+ portfolio founders with thought leaders in the industry, to uncover use cases and opportunities for their respective companies.
Here are some interesting highlights from the discussion:
AI’s evolution from comprehending to transforming data
Broadly, AI comprises two domains: 'Understanding' (a mature field) where AI comprehends diverse data, and 'Modality Transfer' (an evolving sector) involving transforming information between different formats.
As of today, within the 'Modality Transfer' spectrum, text-to-speech technology has made significant strides, generating human-like speech. In contrast, text-to-image, despite being technically mature, faces challenges when applied in production settings. The diagram highlights where our progress stands today:
“Today's text-to-speech applications demonstrate impressive capabilities, including high-fidelity speech generation, adjustment of emotion and tone, real-time accent conversion, and even voice dubbing. These advancements highlight the transformative role of AI in communication and entertainment,” said Ankur, CEO and Co-founder of Murf AI.
Here’s a great sample to bring it to life:
Being business backwards is key
To truly maximise value, you have to think business backward when adopting GenAI into your product or operations. Nitin, the CBO and Co-founder of OfBusiness, gave us an excellent case in point.
Nitin noticed that customers spent a good amount of time engaging with sales reps to stay updated on market trends, pricing, and product news. The sales process was largely dependent on these reps, and any turnover could potentially impact customer relationships.
To tackle this, Nitin thought, why not let customers self-service via a chatbot? He and his team first laid out all the relevant use cases and then rolled up their sleeves, using open-source tools to build the chatbot. Fast forward to today, the chatbot can hold its own in conversations about 10,000 products and 50,000 SKUs, and it's even multilingual, supporting ten languages. Quite a success, considering they chose to automate a traditionally offline process.
Nitin's story reveals two things: 1) By thinking business backwards, you can create creative solutions to tangible problems, even if it’s for an offline process; and 2) With tech increasingly becoming standardised for many use cases, it's entirely possible to create solutions with a modest AI skill set.
What do you need to implement GenAI in production?
The tech stack for text-based scenarios is typically simpler due to its straightforward implementation. Most implementations include an embedding space, libraries like LangChain/LlamaIndex, and a front-end app. It's a go-to choice for quick prototyping and innovation, which is why you see chatbots in about 90% of hackathon projects.
When you compare this to tech stacks needed for other modalities, it starts to get more complex. A founder (currently in stealth) underscored the importance of controllability and composability on top of image generation models to get them ready for production. Controllability gives you precise control over the content you generate, while composability lets you mix and match different elements to get the results you want. Incorporating these elements can be tough, but when done right, your users will see some pretty remarkable results.
Here's an overview of the two stacks that were sketched out during our session:
Putting AI at the core of your business
To truly unlock value for your users, AI should be at the core part of your product and not just a layer on top. This shift in perspective is something all founders should embrace, before starting on the journey to adopt AI.
The next important decision is to determine whether to build or buy AI solutions. If you're in B2B software, taking the open-source route can give you control and adhere to compliance. On the other hand, consumer businesses might find hosted versions more suitable, provided customer data stays safe.
Culture starts from your own team, and thus, it is important to make each employee a power user of AI. Our founders have been experimenting with AI tools across the engineering, product, sales, and marketing teams. Not only has this improved productivity, but it has also let employees realise the potential of AI themselves. Some have even considered prompt engineering courses.
If you are building or working on generative AI solutions, drop us a line at email@example.com.