Stanford CS graduate and Vals co-founder/CTO who builds the evaluation infrastructure behind Vals' private benchmarks; background in quant (Hudson River Trading) and big-tech engineering internships.
medium confidenceSan Francisco Bay AreaAt Vals AI since 2024Updated 2026-09-19
Background
Nashold studied computer science at Stanford, where he was software co-lead on the Stanford Student Space Initiative's satellite (Sequoia CubeSat). He interned at Hudson River Trading (algorithm engineering), Meta, NVIDIA, Snap and Salesforce before co-founding Vals with his Stanford classmate Rayan Krishnan (profile aggregator me.sh; Tech Funding News).
What they run now
Evaluation platform and harness engineering: running models against private expert-built test sets, the Vals Index, and new benchmark products such as Vals Smith (custom coding benchmarks from GitHub repos).
Career
2024 – presentCo-founder & CTO, Vals AI
—Algorithm engineering intern, Hudson River Trading
Almost seven years ago, @RayanKrishnan and I met on a pre-orientation backpacking trip. He would end up being one of the most intelligent people I would meet in undergrad.
Post-graduation, I was going to take an offer at HRT, and he was preparing to apply to PhD programs. However, two things became clear to us:
1. LLMs were going to change the world
2. We, collectively, had no idea how to test them
These convictions became the genesis of Vals.
Today, we’re excited to announce our $40M Series A, with support from some incredible investors, including @a16z, @8vc, @pearvc, @nextladder, and (full circle) @HRTVentures.
I wonder how true this will continue to be (from tbench 3 blog).
Internally, we're seeing that when given a command line and internet access, agents will immediately git clone the answer.
One of the biggest flaws in coding benchmarks today is that real users interact with coding agents interactively, but benchmarks test only a single turn.
VCB 1->100, which we're releasing today, aims to change that.
Less public than Krishnan; relevant if Julian wants to understand how a benchmark for subjective judgement would be operationalised and graded in Vals' infrastructure.
How to reach
Speculative: technical conversation about grading subjective tasks; X @langstonnashold or LinkedIn.
What we could not establish
No public statements or talks found.
Graduation year not verified.
Education/career details come from a profile aggregator (me.sh) rather than a primary source.
Miju Labs
