Anshul Pandey on building a no-code AI platform for accelerating innovation in financial services.
Subscribe: Apple • Android • Spotify • Stitcher • Google • AntennaPod • RSS.
This week’s guest is Anshul Pandey, CTO and co-founder at Accern1, a startup helping financial services companies build and deploy AI applications via a no-code platform. Our conversation focused on the specific challenges of building AI and NLP applications within financial services. As many of our listeners may be aware of, we recently released the results of our second annual NLP Industry Survey, so I also explored the challenges surfaced in that survey with Anshul. We also discussed Responsible AI in the context of the financial services industry.
Anshul Pandey:
Let me take a very simple example, let’s say an article comes out that says, “A JP Morgan analyst has downgraded a particular company’s rating from buy to hold”. That single article will have a very different interpretation, if you go talk to a credit risk analyst, versus a trader, versus someone who’s in another domain within financial services. So I think there is lots of context that goes into the interpretation. Then beyond the context, you also have these specific vocabularies that are very relevant for certain departments and not so relevant for another. I think there’s lots of depth in the amount of NLP research that we have to do supporting these different domains and applications.
Highlights in the video version:
-
Introduction to Anshul Pandey, Co-Founder and CTO of Accern
Accern accelerates AI workflows for Financial Services
What exactly is Accern?
What are the key components of a modern NLP platform or NLP stack?
Model Hub and making sure models work for an application
Modern NLP Platform will have some sort of Model Hub and integration with BI Tools
Are there alternative visualizations for NLP? What kinds of visualizations work well for NLP and text?
What about entity linking and a graph?
Why was it important for Accern to focus on financial services instead of a general purpose NLP platform?
In the future we’ll have general purpose AI in the enterprise
What is your approach to tuning a model for a specific use case?
If I have a data science team in the bank, why don’t I do this myself (build models)?
The public cloud may not be tuned for very specific areas in financial services and cost is a big factor
What is the workflow for fine tuning a model in Accern?
Wide box approach and an adaptive NLP work really well for Financial Services to make recommendations/predictions
What’s the rule of thumb? How much data is needed to fine tune?
Does Accern provide entity resolution?
What’s the impact of this wave of large language models and how will they manifest themselves in industry?
Is your platform being used to generate signal for trading?
What are specific challenges and some technical challenges?
What languages do you support?
What are some of the general things you are doing to manage some of the risks around responsible AI?
Data work is more impactful than model work in some ways
Do you have a workflow around documentation for data and models?
What specific trends in NLP are you most excited about?
Helping financial institutions work together to build a better fraud models.
So what’s your take on multi-modal models? It make sense to invest in that area right?
Do you recommend to an aspiring entrepreneur to go vertical rather than horizontal?
Learn more about Accern by going to Accern.com
Related content:
- A video version of this conversation is available on our YouTube channel.
- “Managing machine learning in the enterprise: Lessons from banking and health care”
- “Taking Low-Code and No-Code Development to the Next Level”
- Yoav Shoham: “Making Large Language Models Smarter”
- Connor Leahy: “Training and Sharing Large Language Models”
- Nikhil Muralidhar: “MLOps Anti-Patterns”
- Pardhu Gunnam and Mars Lan: “Why You Need a Modern Metadata Platform”
- Rumman Chowdury: “The State of Responsible AI”
Subscribe to our Newsletter:
We also publish a popular newsletter where we share highlights from recent episodes, trends in AI / machine learning / data, and a collection of recommendations.
[1] This post and episode are part of a collaboration between Gradient Flow and Accern. See our statement of editorial independence.
[Photo by Businessman Working On Modern Technology from Storyblocks.]