Engineering notes and longer-form blog posts from the Amodo team, newest first.
Follow-up to our memory wiping work: the NVIDIA BlueField-3 DPU and its BMC were large, unwiped pieces of the compute tray. Being Arm rather than x86, the carve-out, cache flushing and memory discovery all change. We found the DPU can be wiped in under 2 hours.
We fitted a passive optical TAP to our inference recomputation prototype. A frame processor reconstructs inference requests from the tapped link and checks all other traffic against a whitelist. It verified 800 requests while the link was flooded with 1 Gbps of noise.
Follow-up to our memory wiping work: testing enterprise U.2 NVMe SSDs and multi-GPU PoSE label generation. We are now limited by PoSE speed rather than SSD speed, and expect ~2.5 hours to wipe an NVL72 rack's storage.
We rebuilt our inference verification prototype around an asynchronous ledger so it can work with passive network TAPs, verify multiple models on a Kubernetes cluster, and measure verifier vs. prover compute. Early results show a 2–8× verifier compute advantage.
AISI is one of the UK's few pieces of international leverage in the AI race. More bets are needed, and AI verification technologies should be the next one.
What tools are available to ensure everyone sticks to the rules on pacing frontier AI development? Written with Lucid Computing.
Hashing network traffic creates a tamper-evident record of what data has transited a link. We benchmark line-rate hashing at 400GbE across CPU and DPU hardware, with an interactive dashboard of the results.
Wiping cluster memory and proving that it's been wiped. Initial experiments with H200s.
Our first attempt at AI inference verification: implementing and testing DiFR on a tray of H100s against example model attacks (append and append-on-trigger), across 37,500 prompt runs.
Recomputation-based schemes for verifying that an AI cluster ran the workloads it declared — covering inference (TOPLOC, Token-DiFR) and pre-training — plus the non-determinism, game theory, and practicalities that make them work.
Wiping cluster memory and proving that it's been wiped. Initial experiments with H200s.
Verification technologies might be useful even if you're not shooting for an international agreement. Here are a few reasons why.
A photo diary of April and May 2026 in the Amodo office — snapshots of the people, the work, and the projects behind the scenes.
Verification is a process, not a one-shot solution. A ladder of technical mechanisms can build trust towards high-stakes coordination over AI — and nuclear verification history shows how the early rungs get climbed.
A weekend spent testing whether frontier AI models can plan how to machine simple parts — where they succeed, and where they still fall short.
Verifying what a compute cluster is doing might involve "tapping" the network — copying network packets for recomputation. These notes explore what it might take to do that.
The case for Sheffield opening an 'embassy' in London — a base to plug Northern founders into the capital's networks, ambition, and energy without leaving the North behind.
We mapped the power delivery systems of AI data centers. Our diagram explores the data available at each stage, measurement techniques and signals of interest for AI security.
A quick test installation of a 10G network tap for AI verification — passive vs active taps, fibre types, and what scales to production speeds.
We used a DPU as a tray-level bandwidth limiter — an extra layer of security for protecting model weights, achieving near line-rate encryption with hardware-offloaded IPsec.
In 2026 we'll double Amodo. This is how we're thinking about it and 5 high-leverage roles to help us do it.
A two week sprint to develop a piston press for isolating mesoporous silica nanoparticles.
A research project for the RAND Corporation. The work identifies potential preparedness measures for challenging biological Threats
A PCB to for measuring power supply ripple rejection.
One way to think of Amodo is as a business hypothesis testing engine. Our hypotheses are unified by using engineering to improve the world.
Amodo's Theses
In the 20th Century we repeatedly see scientific progress driven by engineering progress. We discover the structure of DNA because we can do X-ray crystallography – new engineering enables new science. In the 21st Century scientists no longer have access to engineers. Amodo unblocks researchers by working as their engineering teams – creating the new tools they need to make new discoveries.
Amodo's Theses
Consultancy is the lost model for DeepTech problem discovery. Combine technical talent with real customer problems and you've got an engine for creating new ventures.
Amodo's Theses
Engineers are too often drawn to work on fast cars and fighter jets – this leaves low-hanging fruit in boring industries. By directing ambitious engineers to problems in forgotten industries, we unlock both progress and profits.
Not every project we've pitched has been funded. These are ideas that, for whatever reason, we're not actively working on anymore. That being said, if you think we should be pursue them again, or if you'd like to pursue them yourself then please reach out!
Failure
In early 2024 we applied to the UK's Research Venture Catalyst to start a new research institute working on Physical Sciences for Pandemic Defence. Working with two Oxford University departments we planned to set up a Focused Research Organisation developing engineering, physics, and chemistry solutions to pandemic defence.
Failure
If faced with a pandemic more transmissible than COVID-19, we would need better face masks to suppress the spread. We spent a few weeks pitching this idea, before deciding the timing wasn't right.
We're a high-growth, world-changing startup. We are always open to excellent talent in any role. Please apply if you're an ambitious top performer in any field or discipline, engineering or not.
Explore opportunities
