Time-based GPU partitioning utility that lets multiple users or applications share a single GPU through exclusive, time-limited partitions with automatic expiration. Built with OpenCL and designed to work across macOS, Linux, and Windows.
distributed systems · hpc · low-level engineering
Ojima Abraham
software engineer · researcher · open source contributor
$ whoami > software engineer @ kalshi · open source contributor $ cat interests.txt > distributed systems, GPU computing, CDC, low-level programming $ git log --oneline -3 > a1b2c3d feat: chronos v1.1.0 released on PyPI > d4e5f6g feat: jikan CDC — building in public > g7h8i9j fix: stacked-seds radial photometry pipeline
// open-source
Open Source
Projects I've built and contribute to.
A Python package for stacking faint galaxy images across broadband filters and preparing them for Spectral Energy Distribution analysis. Includes robust stacking, MAD-based error propagation, radial photometry, and publication-ready plots.
Change Data Capture that just works. Built in Rust to stream every insert, update, and delete from your database in real time, reliably, and without infrastructure dependencies, grounded in Lamport and Chandy-Lamport theory.
The WAVE (Wide Architecture Virtual Encoding) toolchain. A vendor-neutral GPU instruction set architecture with assembler, disassembler, and emulator. Targeting NVIDIA, AMD, Intel, and Apple GPUs.
Recent contributions
// publications
Publications & Research
Selected publications spanning mathematics and astrophysics, with the full record available on Google Scholar.
2026
Toward a Universal GPU Instruction Set Architecture: A Cross-Vendor Analysis of Hardware-Invariant Computational Primitives in Parallel Processors
arXiv preprint (cs.DC)
Defines an abstract execution model, register model, memory model, and instruction set derived from hardware-invariant primitives across NVIDIA, AMD, Intel, and Apple GPU architectures.
2025
CECILIA: The Mass-Metallicity Relation of Low-Mass Galaxies at Cosmic Noon
arXiv preprint (arXiv:2512.00162), submitted Nov 2025
Uses ultra-deep JWST/NIRSpec spectroscopy to study low-mass galaxies at z~2-3.
2023
Perspectives Through a Two-Slit Camera
The Mathematical Intelligencer, Volume 45, pp. 299-306 (Springer)
Department of Mathematics, Franklin & Marshall College. Open access under CC BY 4.0.
// the-spec
The WAVE Specification
The WAVE Specification
Wide Architecture Virtual Encoding
A vendor-neutral instruction set architecture for general-purpose GPU computation. Defines an abstract execution model, register model, memory model, structured control flow, and instruction set derived from eleven hardware-invariant primitives identified across NVIDIA, AMD, Intel, and Apple GPU architectures spanning 16 microarchitectures. Follows the thin abstraction principle: defines what a compliant implementation must do, not how. Validated by a reference implementation with 102 passing conformance tests.
Table of Contents
1. Introduction
1.1 Purpose
1.2 Scope
1.3 Design Principles
1.4 Relationship to Other Standards
1.5 Changes from v0.1
2. Execution Model
2.1 Overview
2.2 Thread Hierarchy
2.3 Thread Identifiers
2.4 Core Resources
2.5 Execution Guarantees
2.6 Dispatch
3. Register Model
3.1 General-Purpose Registers
3.2 Sub-Register Access
3.3 Register Pairs
3.4 Special Registers
3.5 Predicate Registers
3.6 Allocation and Occupancy
4. Memory Model
4.1 Memory Spaces
4.2 Local Memory Details
4.3 Device Memory Details
4.4 Memory Ordering
4.5 Atomic Operations
5. Control Flow
5.1 Structured Control Flow
5.2 Uniform Branches
5.3 Divergence and Reconvergence
5.4 Minimum Divergence Depth
5.5 Per-Wave Control Flow State
6. Instruction Set
6.1 Integer Arithmetic
6.2 Bitwise Operations
6.3 Floating-Point (F32)
6.4 Floating-Point (F16)
6.5 Floating-Point (F64, Optional)
6.6 Type Conversion
6.7 Comparison and Select
6.8 Local Memory Operations
6.9 Device Memory Operations
6.10 Atomic Operations
6.11 Wave Operations
6.12 Synchronization
6.13 Control Flow Instructions
6.14 Matrix Multiply-Accumulate (Optional)
7. Capability System
7.1 Required Constants
7.2 Optional Capabilities
7.3 Matrix MMA Parameters
7.4 Query Mechanism
8. Binary Encoding
8.1 Overview
8.2 Base Instruction Format (48-bit)
8.3 Extended Instruction Format (80-bit)
8.4 Opcode Map
9. Conformance
9.1 Required Behavior
9.2 Implementation-Defined Behavior
9.3 Undefined Behavior
9.4 Conformance Testing
A. Full Opcode Table
B. Vendor Mapping Reference
C. Revision History
// press
Press & Coverage
Selected interviews, features, and third-party coverage.
Empowering the Under-Resourced: Nigerian Engineer Builds Tools to Compete on the Global AI Stage
A feature profile on Ojima's open-source work spanning chronos, stacked-seds, and accessible high-performance computing.
Opportunity Funds Scholar Makes Strides in High-Performance Computing, Systems Engineering
Covers Ojima's path from Opportunity Funds scholar to contributions in HPC and systems engineering.
// about
About & Contact
Software engineer at Kalshi. Expert in distributed systems, systems performance, low-level programming, compiler and interpreter design, parallel programming, and high-performance computing. Based in Brooklyn, New York.
Zero dependencies. Just HTML, CSS & JS. Hosted on GitHub Pages.