YEVI Mawuli Péniel Samuel
Who am I?

YEVI Mawuli Péniel Samuel.

Scientific computing, systems, and computational exploration — at the interface of mathematics and physics.

Self-taught student and researcher, I build computational tools and systems to explore problems in mathematics, physics, and scientific computing. I look for demanding technical projects to learn rapidly, contribute tangibly, and deepen my computer science foundations.

  • Scientific computing, parallel architectures, and low-level optimization
  • Local artificial intelligence, sequential models, and embedded systems
  • Frameworks, runtimes, and experimental research tooling
Deep Learning Systems Scientific Computing Numerical Methods Embedded Systems & IoT
Portrait of YEVI Mawuli Péniel Samuel
About

Understanding, representing, and transforming information

I am an Embedded Systems & IoT student at IFRI-UAC, and a self-taught researcher. Computer science is my core domain: I engineer tools and systems to explore fundamental questions across mathematics, physics, and scientific computing.

I currently explore scientific computing, numerical methods, systems programming, and experimentation framework engineering, with an uncompromising focus on rigor, performance, and reproducibility.

Stack

Technologies & Tools

Systems & Compilers
C / C++ (C11) Rust OCaml LLVM IR Linux / Valgrind
AI & Scientific Computing
Python llama.cpp / GGUF Unsloth FastAPI
Runtimes & Tooling
Tauri v2 Electron Git / GitHub
Featured Projects

Research Axes & Technical Prototypes

Bissi - Local-First & 100% Offline AI Agent

Private AI assistant designed for digital equity (West Africa): Gemma 4 E2B model fine-tuned with Unsloth, quantized in Q4_K_M (3.2 GB), run locally via llama.cpp and FastAPI, featuring document ingestion and LaTeX KaTeX rendering on Electron.

Bissi AI agent interface in dark mode

Stack: Gemma 4 E2B, Unsloth, llama.cpp, FastAPI, Electron, Python

GitHub Code · Hugging Face Model

Mercuria - Reproducible Experimentation Notebook

Proof-oriented scientific experimentation environment: custom Python hypothesis formulation, sandboxed execution, and granular delta tracking between expected and observed metrics.

Mercuria reproducible experimentation notebook interface

Stack: Rust, Tauri v2, TypeScript, Vite, Python subprocess

View project

Elementals - Scientific Language Compiled via LLVM

Design of a programming language dedicated to multidimensional computation and physics. Elementals introduces native hypercomplex numbers (quaternions, octonions), a semantic duality between pure mathematical calculations and side-effect processes, and ultra-fast native compilation via LLVM IR and Clang.

quantum_field.el
->> Native hypercomplex numbers & HyperNumber architecture
element QuantumState:
    as Field
    rotation: HyperNumber = 0.5 + 1.2j + 3.4k
    repr = 'State(|ψ⟩)'

->> Pure mathematical transformation
compute_phase(q, energy):
    q.rotation * cos²(energy) + sin²(energy)

Stack: OCaml, Dune, LLVM IR, Clang, C runtime

View project

logariasmós - Scientific Computing Library in C11

Modular scientific core for linear/bilinear algebra, numerical integration, and probability theory, with strict memory management, cmocka unit test suites, and Valgrind profiling.

bilinear.c
// Explicit allocation & integrity tag verification (magic)
matrix_t A = matrix_from(3, 3, (double[]){1, 0, 0, 0, 1, 0, 0, 0, 1});
tensor_t T = tensor_contract(&A, &B, 1, 0);

if (matrix_is_valid(&A) && last_status() == STATUS_OK) {
    matrix_destroy(&A); // 0 memory leak certified by Valgrind
}

Stack: C11, Makefile, cmocka, valgrind, Linux

View project

Automaton Universale - Runtime for Edge AI & Robotics

Automata framework decoupling perception (AI inference), decision (deterministic core), and action (physical actuators and safety guardrails) for critical embedded systems.

controller.py
from automata import Automaton

app = Automaton(model="edge_obstacle_q4.onnx", name="Titan-Rover")

Stack: Python 3.9+, Edge AI, cyber-physical architectures

View project

Graphy - Graph Theory Library

Library for modeling, manipulating, and analyzing complex graphs, with an extensible architecture for optimization and traversal algorithms.

dijkstra.py
from graphy import Digraph, Arc, dijkstra

g = Digraph()
g.add_arc(Arc(u="α", v="β", cost=1.414))
g.add_arc(Arc(u="β", v="γ", cost=2.718))

distances, predecessors = dijkstra(g, source="α")

Stack: Python, graph theory, algorithms

View project