My Journey with LLMs in FinTech-Notes from a Freshman

The Coding Revolution I Never Expected

As a freshman in fintech, I never imagined that my most valuable tool wouldn't be a complex financial algorithm or a specialized trading platform, but something entirely different: Large Language Models.

After stumbling across Harper Reed's insightful blog post on LLM codegen workflows, I've been experimenting with this approach for the past few weeks—and wow, the results have been transformative.

📝 Key Takeaways from Reed's LLM Workflow

Reed breaks down the process into a surprisingly straightforward workflow:

1. Idea Honing: Use conversational LLMs to develop and refine your concept
2. Planning: Create a detailed blueprint with step-by-step implementation plans
3. Execution: Implement the plan with your preferred codegen tool

What strikes me most is how this approach balances structure with flexibility. You're never "flying blind" because you have documentation and a solid plan, but you're also leveraging AI to do the heavy lifting.

💭 My Fintech Implementation Daydreams

I can't help but imagine how this could revolutionize some of our work:

• Risk Assessment Models: What if we could rapidly prototype different approaches to credit scoring algorithms? We could test various models in days rather than weeks.

• Compliance Checking: Imagine creating tools that automatically scan transactions for regulatory compliance. With LLM assistance, we could iterate quickly on the pattern detection logic.

• Customer Financial Dashboards: Building intuitive visualizations of complex financial data is time-consuming. LLMs could help generate the frontend code while we focus on the data architecture.

🧪 My First Experiments

Last week, I tried using this workflow to build a simple expense categorization tool. The process was eerily similar to what Reed described:

1. I brainstormed with Claude to define what my tool should do
2. Created a spec document that outlined the data structures and categorization logic
3. Used the planning approach to break it down into manageable chunks
4. Started executing with a codegen tool

The strangest part? I found myself staring at the screen during "downtime" while the LLM generated code. Just as Reed mentioned, I started using this time to brainstorm my next feature. Productivity doubled!

👨‍💻 The Lonely Coder Conundrum

One concern that resonates with me from Reed's post is the "single-player mode" problem. Coding has always been a collaborative effort in our team, but these LLM workflows feel isolated. I wonder how we might adapt Reed's approach to make it more team-friendly in our fintech environment?

Perhaps we could:

• Set up shared planning sessions where we develop the spec together
• Assign different components to team members, each using LLM assistance
• Create a standard documentation format for sharing LLM-generated code

🔮 Looking Forward

As a freshman in this space, I'm both excited and cautious. Could these tools make traditional coding skills obsolete? I don't think so. If anything, they seem to elevate our work by handling routine coding tasks while we focus on the creative and financial logic.

I'm curious if others in the fintech space are experimenting with similar workflows. Have you tried using LLMs to streamline your development process? What challenges have you encountered?

For now, I'm treating this as a powerful augmentation to my toolkit rather than a replacement. As Reed put it: "It has been super helpful to do small amounts of work in a big code base."

Next on my experiment list: using Reed's "missing tests" prompt to strengthen our transaction processing code. I'll update you on how that goes!