Elite consulting. Free to build anything.

The creative company of Alfredo Metere, PhD, who founded it and owns it outright. Here he builds whatever he decides is worth building, from record-breaking code to patent-pending inventions. It is also an elite technical consultancy for the hardest problems in high-performance computing, AI, agentic AI security and systems architecture.

Start a conversation → NVIDIA Inception Program Proud member of NVIDIA Inception. Selected for NVIDIA’s program for cutting-edge startups.

The company

A creative company. An elite consultancy.

Metere Industries, Inc. answers to no roadmap but its founder’s. It makes what it finds worth making, and it brings the same depth to clients who need hard problems solved. Every consulting engagement is led personally by the founder.

01

Elite Consulting

Senior advisory and hands-on engagements in the fields where we hold real depth, from first assessment to delivered outcome.

02

Creative Work

Software, inventions and products built in-house, on our own terms, because they are worth making.

03

Ventures

Additional lines of business operated by Metere Industries, Inc. under separate trade names.

Expertise

Where we consult.

Systems, hardware and software architecture where science, supercomputing, AI and security meet. Engagements range from focused reviews to long-term technical partnerships.

  • 01

    Cybersecurity for Agentic AI

    Securing AI agents that act on sensitive data: deterministic enforcement, cryptographic boundaries, access control and tamper-evident audit for regulated and mission-critical environments.

  • 02

    HPC Application Porting & Performance

    Porting and optimizing large scientific codes for GPU and exascale systems on NVIDIA, AMD and Intel hardware, with deep expertise in AMD GPU platforms and ROCm/HIP alongside CUDA, MPI, OpenMP, RAJA and Kokkos.

  • 03

    AI Inference & Accelerator Co-Design

    Hardware/software co-design for AI accelerators, LLM inference kernel tuning, and scale-out architecture for large distributed inference.

  • 04

    Hardware Architecture & Performance Modeling

    Hardware-accurate performance projection of real workloads on novel architectures, to guide design decisions before silicon.

  • 05

    Artificial Intelligence & Machine Learning

    Research and development of new AI models, ML for scientific data analysis, and AI-driven data compression for simulation.

  • 06

    Computational Science & Simulation

    Molecular dynamics, fast multipole methods, numerical linear algebra, and computational physics and chemistry at scale.

  • 07

    Technical Due Diligence & Advisory

    Independent, field-expert assessment of technologies, teams and roadmaps for investors and technical leadership.

  • 08

    Supercomputers & Linux Clusters

    Architecture, deployment and operation of HPC systems, from classic Cray machines to modern HPE Cray EX supercomputers and Linux clusters built from the ground up.

Products

Products and inventions.

Proprietary software, patent-pending inventions and record-breaking code. Some of this work is licensed by Metere Industries; some is published openly as research.

Record-breaking code

Codes that set the mark in their space, from GPU eigensolvers and supercomputer ports to real-time fire in a web page.

Proprietary eigensolver

ALMEIGEN

A high-performance eigensolver for NVIDIA GPUs that outperforms NVIDIA’s own cuSOLVER on NVIDIA hardware. On complete symmetric eigendecompositions in double precision, its lead grows with problem size — and it keeps running at scales where established solvers such as ELPA1 abort.

Problem
Complete symmetric eigendecomposition
Platform
NVIDIA GPUs · FP64
Benchmark
Up to 1.49× faster than cuSOLVER syevd
1.49× faster than cuSOLVER at n = 31,000. Ahead from n ≈ 13,000, and widening.
  • ALMEIGEN
  • ELPA1
Speedup over cuSOLVER syevd (higher is better). Complete symmetric eigendecomposition, 1× NVIDIA A100 40 GB, FP64. ELPA1 aborts beyond n = 21,000 (int32 workspace query).
Show data
Matrix order nALMEIGENELPA1
6,0000.70×0.22×
11,0000.96×0.29×
16,0001.14×0.32×
21,0001.31×0.34×
26,0001.43×aborts
31,0001.49×aborts
Published research · arXiv:2511.18674

Low-Rank GEMM

Matrix multiplication via low-rank approximation with FP8 acceleration. It adaptively chooses decomposition method (SVD or randomized SVD) and precision for each matrix and the available hardware, reaching sub-quadratic complexity while keeping hardware-accelerated performance. The gains come from memory-bandwidth optimization, not computational shortcuts.

Read the paper ↗
378
TFLOPS on a single NVIDIA RTX 4090
7.8×
faster than PyTorch FP32 on large matrices
75%
memory savings
N ≥ 10,240
fastest approach, ahead of cuBLAS
First GPU port · LLNL Sierra

SPARC on Sierra

Alfredo Metere wrote the first GPU acceleration of SPARC, a real-space density functional theory code for electronic structure calculations, porting it to the NVIDIA GPUs of the Sierra supercomputer at Lawrence Livermore National Laboratory.

Related paper: J. Chem. Phys. 158, 204117 ↗
60× speedup of the SPARC code on LLNL’s Sierra supercomputer
Web plugin · Real-time fire · Commercial licence

Livefire™

A fire that is actually burning, in less code than a single still photograph of a fire. Livefire does not play back a flame: it computes a live fire in the browser, every frame, so it never loops and never repeats. Flames lean, tear and pinch off, embers lift and cool, and heat shimmer follows the real heat. It ships as one JavaScript module and one canvas tag, and it is tunable live, with temperature, height and heat release reported in real units.

See the live demo ↗
Live
computed every frame, no video, no loop, no seam
~150 KB
lighter than one photo of a fire: zero image, video, sprite or font assets
20
live parameters, eight flame colours and a colour picker
WebGL
holds its frame rate, pauses when out of view

Inventions

Patent-pending technology for signal acquisition, available for licensing and evaluation under NDA.

Signal acquisition · Patent pending

Metere Decimator

The whole dynamic range of a wide signal through one analogue-to-digital converter per channel, delivered at one level to the demodulator, equaliser or baseband that follows. For wireless and wireline receivers, SerDes, radar, instrumentation and sensing, at any sampling rate.

What it does ↗
108 dB
of span from a 75 dB converter, in the example in the application
1
converter per channel, one analogue path
Quiet
every return to high gain placed in a verified-quiet interval, and no sample clipped, in simulation
Any
sampling rate, hertz to gigahertz
Signal acquisition · Patent pending

Zero-Clip Gain Control

A converter clips any sample that arrives above its full scale, and nothing done afterwards can bring that sample back. With Zero-Clip Gain Control, no sample is clipped, for any input within the system’s bandwidth and range. It is a guarantee with a proof behind it, not a low rate of overrange. It complements the Metere Decimator, from audio to radio.

Licensing & evaluation under NDA →
0
samples clipped, for any input within the system’s bandwidth and range
Proof
a guarantee, not a statistical overrange rate
Every
gain change, with the guarantee intact
Audio to RF
verified in simulation on audio and radio signals

The book

A point is a sphere, small enough.

A new foundation for mathematics, written as a book and machine-checked end to end. Free to read, free to verify, on GitHub.

Book · Open access · Lean 4 verified

Quasi-Euclidean Granular Geometry

Calculus, analysis, geometry, topology, probability, information and dynamics on the axiom that a point is a sphere, small enough.

Axiom. A point is a sphere in ℝn of radius ε, where ε is, by definition, smaller than any other quantity under consideration.

Euclid defined a point as “that which has no part”. This book by Alfredo Metere replaces that definition and rebuilds calculus, real analysis, measure theory, geometry, topology, probability, statistics, stochastic calculus, information theory, nonlinear dynamics, computational complexity and physics on it. Here ε is the resolution of the space, not a parameter sent to zero, and the classical theory is recovered from the granular one as its shadow.

1
axiom: a point is a sphere of radius ε
Lean 4
every theorem checked by the Lean kernel
Python
computational companion that asserts every claim it prints
Open
book under CC BY 4.0, code under MIT

Who we work with

Trusted where the stakes are high.

From national security science to the fastest-moving AI companies, our work sits where deep technical rigor matters most.

National laboratory

Lawrence Livermore National Laboratory

A long-standing collaboration, from research staff and computational science roles to ongoing work preparing large-scale scientific codes for some of the world’s most powerful supercomputers.

Industry

SF Bay Area AI startups

We work closely with AI startups across the San Francisco Bay Area on architecture, performance, inference infrastructure and security, at the pace a startup demands.

Ways to engage

  • Expert phone consultationsOn-demand, one-to-one calls for investors, executives and technical teams who need a field expert’s view, fast.
  • Research & technical collaborationsJoint projects with laboratories, universities and companies, from proposal to publication.
  • Advisory retainersOngoing access to senior technical judgment for leadership, boards and investors.
  • Project-based consultingScoped engagements with clear deliverables: porting, optimization, architecture, security reviews.
  • Product & solution developmentDesigning and building software and systems end to end.
  • And much moreIf it is hard, technical and worth doing well, ask us.

Founder

Portrait of Alfredo Metere

Alfredo Metere, PhD

Founder and CEO

Systems, hardware and software architect and scientist with over two decades of experience across national laboratories, research institutes and the semiconductor and AI industries. His work in GPU computing predates CUDA itself: he was already making OpenGL perform numerically significant linear algebra on graphics hardware. He architected the first Linux cluster of the Department of Materials and Environmental Chemistry at Stockholm University, where he earned his PhD, and Zorn at PDC, the first GPU cluster among Sweden’s national supercomputing centers.

He was a core developer of GROMACS, ported Monte Carlo radiation biophysics code for Karolinska, and is equally at home on AMD GPUs and ROCm, Cray systems old and new, and Linux clusters. His work spans GPU porting of physics codes for supercomputers at Lawrence Livermore National Laboratory, customer leadership for the Aurora supercomputer at Intel, performance modeling of novel architectures at Samsung Semiconductor, and AI accelerator co-design at Etched. Along the way he has written several record-breaking codes, including the ALMEIGEN eigensolver and PASYVAT, a GLSL engine that renders 100 million particles in real time. Today he also leads Enclawed, Inc., which he founded to make agentic AI deterministically secure.

Education
PhD, Physical Chemistry — Stockholm University
Research
Peer-reviewed papers in Phys. Rev. E, J. Chem. Phys. and Soft Matter
Invention
Patents and records of invention in hardware, chemistry and materials
Affiliations
IEEE · ACM
Also founder & CEO of Enclawed, Inc. Deterministically secure AI agents — a hardware/software security and compliance layer for agentic AI in regulated, mission-critical environments. enclawed.com ↗

Approach

How we work.

A small, principal-led firm by design — so the expertise you hire is the expertise that does the work.

Principal-led

You work directly with Alfredo Metere, from the first call through delivery. No hand-offs, no layers.

Independent

Wholly owned and unaffiliated with vendors, so recommendations follow the problem, not a product line.

Confidential

Discretion is the default. Client work, and the clients themselves, stay private unless agreed otherwise.

Ventures

More is in the works.

Metere Industries, Inc. also operates other lines of business under separate trade names. These are not publicly disclosed at this time.

Contact

Have a problem worth solving?

Tell us briefly what you are working on. Every inquiry is read personally, and we will reply to discuss whether and how we can help.