Story Highlights
The Company
Informed is an AI-powered loan verification & fraud detection platform trusted by seven of the top ten U.S. auto lenders.
The Challenge
The company's engineering team didn't have time to manage & optimize its complex Lambda environment. The result was wasted cloud costs, degraded performance, & mounting engineering toil.
The Solution
Informed adopted Sedai's autonomous optimization to manage their complex serverless architecture and dramatically cut engineering toil.
The Results
- 900+ hours of toil saved in 3 months
- 26k+ days of latency eliminated across all Lambda workloads
- 100% production compute runs on autopilot
- $7K saved on a single Lambda function
- Zero production incidents caused by Sedai
Building a Cutting-Edge Cloud at Startup Speed
Seven of the top ten U.S. auto lenders trust Informed’s AI platform to verify their loans. Its platform has processed over $350 billion in loan originations and counting. To meet Informed's complex infrastructure requirements, its engineering team runs everything on AWS Lambda functions.
For Rob Berger, Chief Architect, that last part was both a point of pride and a problem waiting to happen. Because Informed’s customers submit loan documents in bursts, Rob made a deliberate call early on to go serverless:
"Going serverless was pretty perfect, since we cycle between usage and no usage throughout a 24-hour period."
Rob rebuilt the company's infrastructure from the ground up on AWS. Each environment runs 306 Lambda functions, many dedicated to running ML inference models, and 37 Step Functions as orchestrators — all replicated across multiple environments.
For a lean engineering team, managing complex infrastructure at that scale is no small feat.
The Lambda Optimization Problem
Lambda optimization is one of those tasks that's easy to deprioritize, especially when you’re working at startup speeds. But when cost becomes impossible to ignore, teams like Rob’s find themselves spending an excessive amount of time on manual optimization.
"In the past, there were no tools to optimize Lambdas for cost & performance without a lot of manual work," Rob said.
The ongoing manual toil of checking configurations, monitoring costs, & tuning functions quickly consumed engineering capacity.
This problem was compounded by ML models running inside Lambdas and exploding their size. At the time, some functions were allocated as much as 10GB of memory — AWS Lambda's maximum limit.
While setting memory high guarantees functions won’t fail, it’s an expensive habit. The more Lambdas are provisioned, the costlier they become to run. But without a way to safely test and tune each function, Rob's team couldn't risk provisioning lower.
The Cost of Manual Optimization
Needing to drastically reduce costs, Rob planned to build an internal tool to handle optimization. But that would mean taking his engineers away from their day-to-day responsibilities just to build it, never mind spending time monitoring the tool continuously after it was implemented.
"We originally planned to write our own performance analysis and do traditional serverless optimization,” he said. “But that takes a lot of work, and you have to do it periodically. My team just didn’t have time for that.”
Building and maintaining that tool would have created a whole new source of toil for a team that was already stretched thin.
Then, right before the team was about to launch that internal tool, Rob came across Sedai.
"When I found Sedai, there was nothing else on the market that I could find that optimized Lambdas the way Sedai does."
The Autonomous Switch with Sedai
With Sedai’s integrated ML that optimizes Lambdas based on real-time behavior, Rob was able to skip the internal tool entirely, and get continuous optimization without any of the work.
However, finding Sedai was only the first step. Because Rob runs production infrastructure supporting hundreds of billions in loan originations, trusting an autonomous system to make changes in that environment wasn't something he took lightly.
“We were definitely hesitant with using Sedai at the beginning,” Rob said. “We were concerned that it would just start flipping things during production and cause production glitches.”
But Sedai's crawl-walk-run approach quickly dispelled these fears.
Before enabling any autonomous actions, Rob & his team began work in Sedai’s Datapilot mode. It provides full visibility into optimization opportunities & recommendations without executing any changes or risking production.
Rob’s engineers could see exactly what Sedai would do before it executed any optimizations.
“Having the ability to see how Sedai optimizes while giving me the control to execute gave me a lot of confidence to move forward with it at startup scale."
Trusting Sedai’s Impact in Production
With the confidence Sedai could accurately identify Lambda functions needing optimization, the team moved to Copilot mode, which reviews and approves changes before Sedai executes them.
Rob’s team watched the recommendations, verified the behavior, & then gradually expanded autonomy within their Lambdas.
"Sedai is completely transparent in what it optimizes. With that insight, we were able to confidently give it full autonomy. We just said, ‘Ok, let’s just turn it on.’"
Give Your Team Time Back
See how Sedai cut 900+ hours of engineering toil for yourself.
Today, 100% of Informed's production compute runs on autopilot, with zero production incidents caused by Sedai.
"We expected a full engineering effort to optimize Lambda functions. But Sedai just handles it for us, on autopilot."
The most remarkable part of Rob’s implementation of Sedai is how unremarkable it was.
"We pretty much got it set up and then ignored it," he said.
Impact that Goes Beyond Budget
By adopting Sedai, Informed has dramatically reduced their cloud costs. But beyond budget, Informed eliminated 900+ hours of engineering toil in three months and 26k+ days of accumulated latency — all without a single production incident.
For an AI startup processing loan documents for the biggest lenders in the country, that performance impact is significant. Faster Lambda execution means faster loan verification, faster fraud detection, & faster decisions for the banks that rely on Informed's platform.
Before Sedai, the team expected to spend engineering time building internal tooling, running test suites, & manually tuning functions on a recurring basis. Lambda optimization would have become a permanent drain on an engineering team that had no time for it.
"Sedai just takes off a whole area of worry,” said Rob. “Normally, we would probably always be looking at our AWS spend and where costs were spiking. We just don't have to do that anymore."
With Lambda optimization toil eliminated, Rob's engineers can focus on what they're actually there to do: building the product.
"My engineers want to learn and build new things, not worry about delivering new features or spending time on optimizations. Sedai handles that with autonomy, so my team can just build."