Beyond Chat Interfaces: How Brightwave’s ML Engineer Thet Naing Envisions the Future of AI in Finance
Thet Naing is a Machine Learning Engineer at Brightwave, where he applies AI workflows to transform financial data processing and analysis. Prior to joining our team, he led System's LLM engineering and information extraction team using novel LLM techniques to build a 50+ service NLP pipeline to expand their scientific knowledge graph by 300x.
About Brightwave: Brightwave is the leading AI-powered research and diligence platform for financial professionals, helping investment teams uncover high-value insights and accelerate decision making with confidence.
In the rapidly evolving landscape of AI, finance professionals are searching for solutions that go beyond basic chat interfaces to deliver meaningful insights from complex data. Following the AI Engineer Summit, Brightwave's Machine Learning Engineer Thet Naing shared some valuable perspectives on how agentic AI systems are transforming financial workflows and why context-aware, adaptable tools are the future of the industry. In this exclusive Q&A, Thet reveals how Brightwave is building AI solutions that anticipate user needs while maintaining transparency and flexibility across different levels of technical expertise.
“Context” and Anticipation: Key Themes from the AI Engineer Summit
Q: Thanks for sitting down with us, Thet! First of all, can you tell us a bit about your experience at the AI Engineer Summit last week and what stood out the most for you?
Thet: It was a great opportunity to meet other AI engineering professionals and learn about how people are using LLMs across a wide variety of both vertical and horizontal applications. The theme of the conference was focused on agentic processes, and it was interesting to see how many similarities there were across the various industry use cases. For me, one thing that really stood out was the focus on context and anticipation of needs.
To dive into that a bit more, by “context” I mean reducing the need to extensively prompt and instruct within the tools that you’re building. For example, in Brightwave, we’re thinking of ways we can better anticipate user needs by building a better understanding of who they are and what they want to do based on the information they’ve provided to us.
Building Trust Through Transparency: Brightwave's Approach to User Engagement
Q: How do you think about building and designing systems that keep the user engaged in the product while taking context into account?
Thet: On a philosophical level, our approach is to try to do as much as we can for the user so we can get them to a 30-second “wow” moment. To do that, we not only have to show them the work being done, but we also have to make sure that once the work is completed, the user has the ability to go in and see the very detailed specifics about how we came to our conclusions for a particular answer or set of calculations.
This includes building out really robust methodologies for attributing information to specific places and documents, having trustworthy sentence-level citations and being able to click into and decompose any problems that come up. We want the user to always understand the steps that we took to get them to a particular conclusion.
Understanding Agentic Systems: From Hard-Coded Workflows to Autonomous Problem-Solving
Q: We’ve been hearing a lot of buzz lately about “agentic workflows.” What does that mean to you and how is it relevant to Brightwave?
Thet: Agents are essentially systems built around LLMs that have access to tools, memory and planning capabilities. These tools can be used for searching particular types of documents, taking some sort of action that produces an outcome, or performing a calculation and turning the information into a chart or a graph. A lot of the focus around agents is being able to build systems around LLMs that are able to autonomously or semi-autonomously arrive at the answer or solution to a problem without each step being explicitly programmed.
The earlier paradigm around workflows involved a series of explicit requests to LLMs. For example, we would ask an LLM for an answer, then ask it to execute this tool, or perform another calculation. Now it’s a lot more flexible and not so hardcoded. The steps aren’t explicitly programmed in that order and agentic systems are able to decide for themselves: does it make sense based on the user’s request to perform an EBITDA calculation or to draw up a certain chart? Does it make sense to terminate a process after not finding relevant information, or does it make sense to structure relevant information into a report that consists of three paragraphs and a table? This flexibility allows us to give much more tailored experiences to our users and build systems that can adapt to their needs.
Emerging Trends in AI and Finance: Insights from the Summit
Q: That’s helpful context. Pivoting to one of the themes of the conference, which was around AI and Finance, were there any emerging trends or new developments that caught your attention?
Thet: To me it’s becoming increasingly clear that a simple chat interface isn’t going to be enough, both within and outside of Finance applications. The future of AI will require a lot more than chat to be successful in specific workflow applications. There needs to be an emphasis on meeting users where they’re at and integrating into their existing workflows and tools while building systems that are tailored and curated for their needs.
At the same time, [ChatGPT’s] Deep Research has shown that users are willing to wait longer for LLMs to do their work. But that wait time has to be offset by some meaningful outcome that is appropriate for the time it took. This goes back to the idea of anticipating user needs — instead of choosing which model you want or how long you want to wait, there seems to be a shift towards telling the LLM what you want and having the system figure it out for you.
Dynamic Data Processing: Brightwave's Multi-Modal Approach to Information Analysis
Q: How does Brightwave fit into this picture? Is there a particular feature or approach you’re most excited about?
Thet: I’m excited about our ability to handle all types of data, whether it’s PDFs and Word documents or structured Excel spreadsheets, and then being able to dynamically decide the right output for the user. This means no longer just using a chat interface or outputting simple text like many of the systems out there today do. It’s being able to determine that certain types of content make the most sense. This may be plain text, it may be tables or charts that users can then export, or it may even be a combination of all of these.
Building out this more richly typed interface makes it more valuable for users in their workflows and the different types of tools they already use. So for example, being able to upload many Excel documents, run analyses over them, getting the outputs in a structured tabular format and being able to export that back to Excel down the road. Or creating graphs that automatically populate into a PowerPoint deck. These types of very workflow-centric things focus on the users we’re trying to serve, and that’s the vector that I’m most excited about.
A Pragmatic View: When Agentic Systems Make Sense (and When They Don't)
Q: Let’s talk hot takes. What’s a bold opinion you have about “agentic” systems in AI, especially as they apply to finance?
Thet: My biggest hot take is that not everything is — or needs to be — an agent. A lot of people are using the term “agent” when they’re simply using an LLM. At the same time, not everything needs to have a 20-step dynamic workflow. Sometimes it’s just about knowing what your customers want and need. I think at times people lean towards agents because they think that it’s a catch-all for solving problems, but ultimately, the solution you should try to build is the thing that solves your user’s problem.
Q: Before we wrap up, is there anything else you’d like to share with those curious about where Brightwave is headed?
Thet: One thing we’re focused on is supporting different user preferences and skill levels. Some users want an almost “hands-off” approach where the AI handles everything. Others want more control to be able to specify each step or validate each tool the AI uses. There’s a sliding scale of different users with different appetites for how much input they want, and I think it’s important to build systems that can account for all of these use cases. We’re designing Brightwave with that in mind, and that flexibility really matters in finance, where you might have one person quickly scanning data and another doing deep, detailed analyses.
Key Takeaways:
- AI in finance is evolving beyond chat interfaces toward contextual, workflow-integrated solutions
- Effective agentic workflows in financial applications require careful design around user needs
- Financial professionals seek varying levels of AI automation based on their technical comfort and task complexity
- Multi-modal data processing capabilities are essential for comprehensive financial analysis
Get in Touch With Brightwave Today
Want to learn more about Brightwave? Reach out to us via our contact form to get in touch or schedule a demo.
