SimulationMaxxing: How we ship agents 20× faster — Aman Gupta (Nubank) + Shreya Rajpal (Snowglobe)

AI Engineer · 16 min · 143 sentences · from YouTube's caption track

Each timecode opens YouTube at the start of that sentence. Line anchors (#s42) are the cue ids in the WebVTT, and every line carries its start and end seconds. All transcripts has every talk, and the whole corpus as one file.

00:01[music]
00:12Hi everybody, my name is Shrea.
00:14I am the CEO of Snow Globe and we have with us Ammon who is a principal machine learning engineer at new.
00:21And this talk is going to be about simulation maxing and how you can ship or how new bank ships uh agents 20x faster using simulations.
00:32Hey everyone, I'm Aman.
00:34Uh so let me talk about New Bank at a glance.
00:37We are the lead leading digital bank in Latin America.
00:40We have 135 million customers uh in Brazil, Mexico, and Colombia and we are launching in the US real soon.
00:47uh our quarterly revenue crossed 5 billion in Q1 2026.
00:51Our NPS customer love is very high and we are the perfect company for using AI agents for customer support where human Xpers are fanatical customer care people and AI agents together solve customer issues in a fast empathetic and correct manner.
01:08AI handles a lot of routines end to end.
01:11uh humans focus on the hardest and longtail cases and together we aim to delight our customers.
01:19So as Shria said this talk is only about uh one thing really.
01:24If you generate your eval data in sim instead of waiting on production data you can ship agents 20x faster and we'll give you evidence for that.
01:34So let's start with the results directly.
01:36Uh uh so this is you know the average of TNPS which is a measure of customer satisfaction for five of our AI agents in production and at the beginning uh they were not so great but now in a with a few months of
01:51uh of work and a few quarters worth of effort we've been able to massively increase the TNPS and customer love for our AI agents and many of them are approaching human quality and this data is a bit stale
02:03many of them are exceeding human quality.
02:05So we are at the stage where we are actually able to show proof that this actually works in production.
02:11Uh here's a QR code for our KD paper in case you want to check it out.
02:14It's going to be presented uh in Korea uh in August.
02:21Uh awesome.
02:21So we open with results and you know it's really about this journey of how do you implement the right systems uh for evaluation in order to you know be able to achieve those results.
02:31So in this talk we basically split it up into these three sections.
02:34The first is why evals are so important and essential but why they're also the bottleneck from being able to you know do a lot of high throughput experimentation to get the results that Amen was showing earlier
02:44and then why the second part of this talk is about why simulated data uh works as well as collected data and helps you circumvent a lot of this bottleneck that we're going to talk about and then the third part is really digging deep into the systems and the findings that we had by implementing
03:00uh you know this framework at scale in new Uh, awesome.
03:05So, eval there this there's like a whole talk track dedicated at this conference to eval.
03:10Evals are absolutely critical for building good agents and evals are really only about two things, right?
03:16There's metrics and there's data.
03:19uh metrics again I hope you attended many of the amazing talks yesterday on the evals track but metrics are really while they're challenging we have a playbook for how to build metrics that are really well aligned with you know the rubrics that we care about right which is you essentially use um you know LM as a judge
03:35style uh you know classifiers and you align it with human judgment and and getting human data and you can iteratively build on it using you know auto optimization and autoprompt tuning uh techniques
03:46the thing that's a bottleneck And that still remains very challenging and unsolved is what is the data that you're actually computing these metrics on and that process is very timeconuming and very expensive
03:59specifically so for agents.
04:01So once again people have been talking about you know if if you're around like machine learning era circa you know 2018 it would be like ML work is 85% data work right so data has always been challenging but with agents the level of sophistication
04:14that data requires is just so much more expensive.
04:17So here's examples you know of what structured ML data looked like uh and what even early era of AI uh data for you know chat bots or single turn QA which was you know much more manageable and tractable and you could still think of it as these structured rows.
04:32Uh but now for multi-turn agents each data point is a trajectory with a lot of internal tool calls that all need state etc. And that becomes very even one data point is expensive to generate and annotate right but scaling it up on a sample set becomes even harder.
04:49Um so how teams primarily get eval data today is two methods.
04:54Uh one is either manual or the second is production traces and both of those are painful but for different reasons right manual authoring I mean this one's obvious it is so timeconuming
05:05especially the kind of data that I was showing you on the previous slide.
05:08You have to hand plan you know state updates.
05:10You have to hand plan the trajectory that you want to take this uh you know agent data point in and synthetic uh the the state for these data points has to stay consistent across all tool calls etc that you make right production traces in comparison are almost free.
05:26You don't have to pay for them.
05:27You're going to get them anyway.
05:28But the cost is that you're testing on real live users every time you're testing it.
05:33So for example, if you want to run, you know, many parallel experiments at scale, which is often something you want to do in a machine learning setting is it becomes very hard to do that in prod.
05:46So for those of you who build agents, you know how difficult building a new agent and maintaining uh an existing agent is.
05:53So if you want to change the agent harness, you want to change the prompt, the tools, anything around the the LLM that powers your agent, that can take a few hours and in in the best case.
06:02But as Fria mentioned, running offline evals on hand curated data can take a few days.
06:08And then the the worst part is if you run an AB test in production and monitor regressions, see if you get a statistically significant improvement of of a previous version
06:17that can take forever because customer feedback is through some kind of a feedback form and it can be sparse, it can be noisy and can take forever to know am I moving in the right direction or not.
06:29So I'm here to tell you that uh we have verified in production that yes simulations circuit this timeline short circuit this timeline from a few weeks you can go to less than a day sometimes even a few hours a few minutes
06:41and uh as I mentioned earlier uh you know with simulations it takes a few hours uh without simulation it takes a few hours to change the harness and offline evals
06:50and all of that's become super fast now with simulation the agent harness chain still takes a few hours running offline evals becomes a lot easier because you have a simulated user to fall back on.
07:02You're not waiting on hand curated data.
07:04You're not waiting on you know people to create data for you.
07:07And the best part if you want to know that am I doing well in production?
07:12Is this version going to do well?
07:14That can take just a few hours.
07:15So this has been a real gamecher for us at New Bank.
07:20Um all right.
07:21So we talked about you know like simulations shortening the release cycle but what is mechanically a simulation you know it's existed in robotics it's existed in you know a lot of different domains but simulations for agent eval you know mean a specific thing and this is really where snow globe enters and you know sets up a lot
07:39of that core infrastructure for you.
07:40So simulations are essentially you know it works by first like wrapping your agent you know using a um uh like pointing the snow globe SDK at your agent without requiring any code changes and you know you can gro what are the tools that you really need in to mock in order to make the simulation really work.
07:57The second input is about how you want to drive or steer your your simulation.
08:02So you know who are the personas what are the use cases what are the data points that we're really simulating.
08:07Once you have those inputs ready, snow globe essentially runs the simulation and on the output you end up getting you know first just a data set.
08:15So you get like thousands of multi-turn conversations against your real agents.
08:19These conversations have you know tools mocked etc. have all of the bells and whistles needed to be actually you know useful for evaluating your agent already ready and then you run like judges on top of it so that on this data set you get you know per turn information about how your agent is really behaving and it
08:36then like pipes directly into you know your evaluation pipeline so that you can really get data on demand.
08:42Um this is an example of what a sim what the what what a concrete output of simulation is like right so this is as an example like one simulated persona and then the corresponding conversation or data point for that simulated persona
08:56so here for example this is totally synthetic a user named Maria Souza uh you know she wants to order a credit card this is who she is she's 34 a designer first-time credit card customer
09:07in addition to you know this basic information about this synthetic user we also will generate, you know, this essential grounding data as an example, this fake address, a fake credit card, etc. that stays consistent
09:20across your agent execution so that you know your agent can actually be evaluated against that.
09:24And then you also have things like tone, voice, how do they really interact?
09:28And then oops, and then when you actually run this persona through the conver through the agent, you end up getting, you know, this uh conversation that's consistent with the simulated intent, right?
09:39So you see for example that this user has very curt oneline messages in terms of how they interact because this is how the persona was set up or that you know when
09:48the agent wants to when the agent runs its tool when Maria enters the chat the agent is actually able to get a valid output mocked by the simulator that you know is able to identify who the user is where the address is the address can be verified
10:03etc. Um and then finally all of this is also grounded in like account data as well.
10:10So for those of us building real agents, you know how hard things are.
10:15Uh you don't know if your agent is good, you don't know if your evals are good.
10:18You have no idea what's happening.
10:19And that was me a few quarters ago.
10:21Uh since the theme of uh this year's conference is self-improving agents.
10:26So we are also kind of jumping on the bandwagon.
10:29So this is our self-improvement loop.
10:31You ship your agent, you observe, you create robust evals.
10:35You spend a lot of time creating robust evals using automated prompt optimization with algorithms like Japa.
10:41And then you also do simulation.
10:43You run simulation and you pipe the simulation data and real data through your evals.
10:49And that gives you rich signal to then optimize your agent harness.
10:53And once you verify the agent harness optimization is good, you ship.
10:57And this has enabled us to do many many many AB tests for a given agent.
11:01And we have many many such agents across the company.
11:04So this has been a real gamecher for us in terms of improving our agents and the proof as I already showed you is that you know the quality of our agents is going up uh pretty dramatically.
11:14This is some data on how simulation really tracks real production data.
11:18So when you run uh simulations you want to compare is my simulation comparable to production data.
11:25So we looked at sim quality via evals.
11:28So far uh you know quality of uh with evals from sim and real data the correlation is pretty high and we had human review done where 80% of our domain expert labels confirmed
11:39that sims give us usable data not just for uh mature agents but also for green field agents.
11:45So this has been uh pretty useful for us in bringing up a new agent from scratch and also taking an existing agent and and massively improving it.
11:55And as as we as I mentioned, you know, simulation is not just about improving uh and sh also talked about this.
12:00We are derisking a lot of things that could have made to production.
12:04For instance, we caught a regression uh with simulation that could have made it to production, but simulation caught it.
12:09And at the same time, we also caught in another agent an issue uh which could have lowered our self-service rate making it less efficient.
12:16And all of this has resulted in pretty large improvements.
12:20Uh for instance, one of our agents, the TNPS, has 2xed uh thanks to simulation, robust evals, and investing in something super principled rather than, you know, throwing something at the wall and see what sticks.
12:33Um it's been pretty awesome to see simulation saving us so much time.
12:37So now the team uses snow globe and tries different ideas and they're able to short circuit launching AB test.
12:45They don't launch until they're happy with the same output.
12:47So imagine if you're running 10 AB tests a quarter for your agent or probably more if you have more data.
12:53Now you can just run a bunch of stuff through SIM and launch just one AB test which kind of shortcircuits launching the first five or six.
13:01And at the same time uh this this is often discussed in customer service circles that TNPS can come at the cost of self-service rate.
13:08Sometimes there's a trade-off but I'm happy to report that we are not compromising all self- service rate with SIM.
13:14We are also able to pretty dramatically improve SSR across the board.
13:18But in one particular case, we were able to improve it by 4%.
13:22All of this makes us a much more efficient company and makes sure that we keep our customers really happy and we keep our SSR rates and TPS rates very very high.
13:33This is one example that I am most excited about.
13:35As you know, all of you know open source models have been on a tear recently.
13:40So many of them have come out.
13:42So, we've been trying to test various open source models in our agent harnesses and see which ones are the best.
13:48And with SIM, we have the perfect recipe to just throw a bunch of ideas, open source models at our agent harness and eval setup and see which model versions
13:59uh you know really really work for us.
14:02Uh this has saved us multiple weeks of effort and probably underelling it.
14:06It's it's been it's been pretty awesome and uh we have AB test running as I speak uh figuring out are these open source models as good as you know uh the frontier models.
14:17So SIM is is is been super super helpful in helping us figure out uh kind of you know spreading our cards uh a bit.
14:26Um awesome.
14:27So um the core three takeaways from this talk right is about um where eval is today and how you can really remove a lot of bottlenecks to it.
14:36So we lied when we said earlier that there's just one thing you should take away from it.
14:40That one thing is still essential but there's a few key downstream things that you can unlock if you know you adopt it which is the first is if you generate your evaluation data in simulation
14:50rather than solely relying on production to get signal on how you know different agents are performing uh you'll be able to undercut or you'll be able to short circuit a lot of the uh bottleneck in in releasing you know versions of your agents much faster.
15:04The second is in order for any of these gains to really be unlocked uh you know you really need to close out the sim toreal gap.
15:11So you need to you know set up like offline online human review kind of metrics to really understand how sim performs visav real production data that you've seen uh so that you you you are able to kind of like trust the results of these simulations.
15:25Um and the third is again there's so much excitement around you know like auto research self-improving agents RSI etc. uh in an enterprise setting when you're building an agent, it really does come down to two things, data and metrics.
15:38If you have align metrics that are able to really catch the signals you care about and you have a reliable way of generating data that those metrics can give you signal on, it's then very easy to put together a loop of an agent that you know continuously improves itself from like feedback it receives from all of these
15:55places.
15:56Uh which you know again is like where the future of this field is heading.
16:01Um, awesome.
16:02That's that's us.
16:03Uh, thank you for coming for the talk.
16:04Uh, there's few like contact links and details for us, you know, if you guys want to get in touch and otherwise we'll be around here if you guys want to ask us any questions.
16:27[music]