At FurtherAI, we focus on solving the hard, overlooked problems inside insurance workflows. These are the tasks that drain hours, introduce risk, and require precision. Our engineering team builds systems that handle this complexity reliably and at scale. One of the people doing that work is Frieda Huang.

Why FurtherAI?

Frieda knew she wanted to work at an AI startup where the technology directly improves people’s day-to-day work. Insurance teams deal with long documents, inconsistent formats, and information that is difficult for AI to interpret. These challenges are not flashy, but solving them has real impact.

“I wanted to build AI agents that actually help people,” she says. “The work underwriters and brokers do is important, and if we can make it easier and more accurate, that matters.”

What excites you most about working at a startup in insurance tech?

What excites Frieda most about insurance tech is the opportunity to rethink how agentic systems operate in environments with high document volume and limited structured ground truth.

Much of her work today centers on:

• making the platform more agentic across workflows like submission intake and policy check
• building evaluation frameworks that measure extraction accuracy
• enabling users to provide ground-truth feedback that improves the system over time
  

“This space pushes you to think carefully about reliability,” she explains. “There’s a lot of nuance in the data, and getting it right is the core of the product.”

What’s a recent technical challenge you worked on, and how did you tackle it?

One of Frieda’s recent projects involved aligning similar sections across multiple policy documents, each more than 300 pages. Policies use different formats, labels, and structures, making direct matching extremely difficult.

Frieda designed a system that uses semantic similarity and the Hungarian algorithm to align endorsements and coverage items, even when the formatting or values differ. A second clustering layer groups only items that are directly similar, which prevents incorrect groupings.

This approach handles real-world document inconsistencies and helps brokers identify changes and gaps far more quickly than manual review.

Who or what inspires your approach to engineering?

Frieda is motivated by people who take a first-principles approach to engineering and focus on long-term impact over shortcuts. She values clear thinking, optimism about the future, and the discipline to build things the right way.

What are you most excited to build at FurtherAI next?

Looking ahead, Frieda is focused on strengthening the agentic scaffolding across the product. Her work includes developing the systems that allow LLMs to observe, act, and ask for human guidance when needed. She’s also building evaluation tools that help the team measure accuracy internally and enable users to refine model outputs themselves.

“This feels like an early-moment platform shift,” she says. “There’s so much to build, and the opportunity to shape how these systems work is huge.”

Frieda brings consistency, care, and a high bar to her work. She goes beyond expectations to make sure what we ship is reliable for the teams who depend on it. We’re lucky to have her building here.

If you’re interested in solving hard problems with a team that cares deeply about quality and impact, we’re hiring. Link to open roles here.