Computer engineering · ML hardware · scientific instrumentation
Computer engineer working across machine-learning hardware, ASIC/FPGA design, scientific instrumentation, and adaptive on-chip learning.
I build hardware/software systems for machine learning under physical constraints: latency, power, numerical precision, memory movement, timing closure, radiation, and deployment on real devices. My work spans algorithms, software tooling, HLS/RTL implementation, verification, FPGA/ASIC workflows, and documentation for reproducible scientific use.
Technical focus
Areas I work across
What I build
Projects at the algorithm/hardware boundary
Hardware-aware ML systems
Tools and workflows that translate machine-learning models into efficient FPGA/ASIC implementations.
- hls4ml
- hls4ml-trainable
- ENABOL
- quantization
- fixed-point deployment
Scientific instrumentation hardware
Low-latency and reliable digital systems for experimental physics and detector readout.
- Fermilab
- CERN
- particle detectors
- radiation tolerance
- ASIC workflows
Adaptive and neuromorphic learning
Learning systems based on local adaptation, energy constraints, and on-device plasticity.
- NRCSTK
- local learning
- metabolic constraints
- edge adaptation
- non-backprop learning
Computer architecture and reliability
Reliability-aware architecture work across timing, memory systems, and resilient execution.
- AMD RAS
- CPU pipelines
- memory hierarchy
- ECC
- aging
- fault tolerance
Full-stack research engineering
Bridging algorithms, software, hardware implementation, experiments, and documentation.
- Python
- C++
- HLS
- RTL
- verification
- reproducible workflows
TL;DR
How I got here
My path started in robotics engineering in Barcelona, where I learned to think across physical systems, electronics, control, and software. In Brazil, I moved into applied physics and industrial AI, working on deep-learning methods for imaging and reservoir characterization. During my Ph.D. at Northwestern, that background evolved into hardware-oriented machine learning: low-latency inference, ASIC and FPGA implementation, particle-detector instrumentation, and adaptive learning systems that can operate under tight physical constraints.
Across these stages, the recurring theme has been the same: I like systems where algorithms have to survive contact with physics. That means numerical precision, latency, power, timing closure, radiation, memory movement, and the engineering details that decide whether an idea can actually run.
Trajectory
Evidence graph
Skills tied to projects and environments
Contributions
Concrete work behind the profile
hls4ml original author and contributor
Helped build the open-source workflow that translates machine-learning models into FPGA/ASIC implementations for low-latency scientific applications.
hls4ml-trainable creator
Developing trainable extensions and hardware-aware learning workflows for hls4ml.
ENABOL creator
Building adaptive training controllers and experimentation tools for on-chip learning and edge-AI systems.
NRCSTK neuromorphic research
Research on local, energy-constrained learning systems and adaptive neural dynamics for on-device plasticity.
Radiation-hard in-pixel AI ASICs
Digital logic and layout for pixel readout chips with neural-network classifiers, radiation-hard design patterns, and sub-10 ns inference targets.
Reconfigurable neural-network ASICs
Architecture, RTL, verification, radiation-hardening, and tapeout work for TSMC 65 nm detector-data compression accelerators.
Cryogenic ML and readout ASICs
GF 22 nm FD-SOI cryogenic accelerators, quantum-control/readout systems, and mixed-signal digital control for low-temperature operation.
Wolf EDA automation
Tooling for digital implementation flows, including synthesis, place-and-route, extraction, and signoff workflows.
Scientific detector AI publications
Published work on in-pixel AI, smart pixels, cryogenic readout electronics, detector compression, and hls4ml-based acceleration.
Industrial imaging and geoscience AI
Deep-learning methods for borehole image analysis, reservoir characterization, fracture/breakout detection, and uncertainty estimation.
Patent contributions
Patent applications for ultrasonic image artifact removal, automatic breakout detection in reservoir well images, and hardware-measurement-based analysis methods.
Personal research tooling
Reusable Python tools including Nodus for job/workflow orchestration and Pergamos for dynamic HTML reporting.