Kencollin — Systems & Machine Learning Engineer

Engineering Philosophy

My work revolves around solving complex technical bottlenecks at the intersection of high-level software abstractions and physical execution. Whether optimizing LLM token throughput on CUDA devices, deploying multi-tenant mail architectures, maintaining national e-learning devices in public sector sub-county offices, or simulating stochastic option matrices—I prioritize mathematical rigor, deterministic execution, and operational reliability.

Low-Level Execution

Building memory-safe binaries with Rust and PyO3 bindings for zero-cost abstraction and performance-critical pipelines.

Scalable Infrastructure

Deploying production mail networks, edge computer vision models, and robust network diagnostic environments.

Key Highlights

ML Evaluation & Serving: Optimized custom LLM inference pipeline achieving low-latency batch processing.
Ministry Leadership: Spearheaded digital literacy device maintenance & network setup across sub-county schools.
Quantitative Systems: Implemented Heston & CIR stochastic volatility pricing engines in Rust and Python.

Explore full career trajectory

Primary Technical Domains

Machine Learning & Computer Vision

LLM inference optimization, custom CUDA kernels, model evaluation, serve architectures (vLLM, TensorRT), and edge vision models for constrained hardware environments.

Public Sector Infrastructure & Systems

Field network provisioning, Digital Literacy Programme (DLP) device management, multi-tenant mail architectures (WigoSend), and edge hardware diagnostics (SDR).

Financial Econometrics & Rust Tooling

Binomial Trees, Heston Stochastic Volatility calibration, Merton Jump-Diffusion models, combined with high-speed Rust PyO3 desktop & CLI software design.

Academic Qualifications

Formal education in software engineering and financial engineering.

Sep 2024 – Present

Master of Science in Financial Engineering (MScFE)

WorldQuant University — Remote (USA)

Graduate study in stochastic modelling, derivatives pricing, and algorithmic trading, with project work in Heston/Bates calibration, Monte Carlo repricing, and delta hedging.

2016 – Nov 2022

Bachelor of Science in Software Engineering

Murang'a University of Science and Technology — Kenya

Foundation in software engineering, systems programming, and applied machine learning. GPA 3.5.

Engineering Experience & Leadership

Demonstrated history across Machine Learning engineering, Public Sector Tech Leadership, and AI Evaluation.

Machine Learning & Systems Engineer

AI Optimization & Serving Architectures

Present

Architected high-efficiency LLM inference pipelines using vLLM and TensorRT-LLM, optimizing token latency and memory bandwidth utilization.
Engineered custom CUDA kernels for specialized matrix operations and attention layer acceleration in resource-constrained environments.
Designed robust evaluation suites for complex LLM task benchmarks, benchmarking model hallucination rate, throughput metrics, and context window drift.
Constructed containerized serving endpoints backed by dynamic batching and load balancing to sustain high concurrency workloads.

TensorFlow PyTorch CUDA vLLM Docker TensorRT

ICT Officer & Digital Infrastructure Lead

Ministry of Education — Sub-County Office

Public Sector Engagement

Spearheaded digital infrastructure operations across sub-county educational institutions, supporting national Digital Literacy Programme (DLP) hardware rollouts.
Configured, repaired, and managed hundreds of student e-learning tablets, teacher laptops, and school server units under demanding field conditions.
Designed and executed sub-county local area networks (LAN), Wi-Fi topology, security firewalls, and server rack installations for government offices.
Automated internal administrative workflows and email notification dispatches, drastically cutting document turnaround times for school administrators.

Network Routing DLP Hardware Maintenance Linux Administration Workflow Automation

Task Annotator & Egocentric Video Reviewer

Ego-OTS Computer Vision Project

Specialized AI QA

Audited high-dimensional first-person (egocentric) video streams for the Ego-OTS research initiative, establishing precise temporal action boundaries.
Annotated complex multi-step human subgoals, hand-object interactions, and spatial-temporal bounding boxes for advanced vision-language models.
Maintained strict quality control metrics and benchmark compliance across thousands of video frame sequences to ensure high-fidelity training data.

Ego-OTS Temporal Segmentation Vision QA Dataset Auditing

Expert AI Model Trainer & Evaluator

Populii AI — Remote

Jun 2023 – Apr 2024

Trained and evaluated AI models, assessing accuracy, bias, and efficiency across computer science, coding, and data science tasks.
Ran fine-tuning and evaluation cycles to improve model performance, working with cross-functional teams to refine model training processes.
Wrote and reviewed prompts in computer science, coding, data science, physics, and chemistry.
Performed model jailbreaking and red-teaming across domains to confirm models stay within ethical boundaries.

Model Training Fine-Tuning Model Evaluation Red-Teaming

Prompt Engineer

Outlier (formerly Remotasks) — Remote

Nov 2022 – Jan 2024

Designed and refined prompts to optimize natural language processing models and produce high-quality AI-driven content.
Debugged and refined machine learning models using user feedback and performance metrics.
Applied machine learning principles to prompt design, collaborating with cross-functional teams to improve AI usability and performance.

Prompt Engineering NLP Model Debugging Feedback Loops

Engineering Projects & Case Studies

Detailed system architectures, production software, and edge toolsets.

Desktop & Systems Software

Hybrid Rust/Python Desktop App Ecosystem

Cross-platform high-performance desktop tools leveraging Rust for zero-cost thread safety and math calculation engines, exposed directly to Python GUIs (PyQt/PySide) via PyO3 and Maturin bindings.

Key Innovation: Zero-copy memory sharing between Rust native libraries and Python NumPy array buffers.

Rust PyO3 / Maturin Python PyQt Rayon Concurrency

Quantitative Finance

Quantitative Econometrics & Options Pricing Engine

A modular computational finance suite implementing Cox-Ross-Rubinstein Binomial trees, Heston Stochastic Volatility calibration, Merton Jump-Diffusion models, and Cox-Ingersoll-Ross (CIR) interest rate term structures.

Quant Features: Fast Fourier Transform (FFT) option pricing, Monte Carlo GPU paths, and yield curve bootstrap solvers.

Python SciPy Rust Math Stochastic Calculus Monte Carlo

Hardware & Embedded Systems

SDR Spectrum Mapper & PSU Diagnostics Suite

A hardware diagnostic and RF spectrum analyzer tool built for RTL2832U Software Defined Radio dongles combined with oscilloscope diagnostic procedures for desktop Power Supply Unit (PSU) refurbishing. Maps local RF interference and validates power rail stability.

RTL-SDR GNU Radio DSP / FFT PSU Diagnostic Protocols

Technical Competency Matrix

Categorized skills spanning low-level code, AI models, networking, and quantitative finance.

Rust Programming Advanced

PyO3 bindings, Rayon concurrency, memory safety, cargo build systems, zero-cost abstractions.

LLM Serving & Evaluation Expert

vLLM, TensorRT-LLM, SGLang, PyTorch execution, batch processing optimization, CUDA acceleration.

Network & E-Learning Infrastructure Production

Ministry DLP tablet maintenance, sub-county LAN topology, Mikrotik/Cisco routing, Linux administration.

Stochastic Financial Modeling Proficient

Heston Stochastic Volatility, Binomial Option Trees, CIR Term Structure Calibration, Merton Jump-Diffusion, Markov Decision Processes.

ML Frameworks & Tooling Advanced

TensorFlow, PyTorch, EasyOCR, model quantization, fine-tuning, and evaluation of computer vision and language models.

TensorFlow PyTorch EasyOCR NumPy

Prompt Engineering Professional

Prompt design and refinement for NLP models across coding, data science, physics, and chemistry, plus jailbreak testing.

Fine-Tuning & Model Evaluation Professional

Model training and evaluation, SFT data work, and rating model accuracy, bias, and efficiency.

Python / PyO3 Interop Advanced

NumPy C-API, FastAPI, PyQt5/6, Maturin binary packaging, SciPy optimization routines.

SDR & Hardware Diagnostics Hands-on

RTL2832U spectrum mapping, GNU Radio, ATX Power Supply Unit refurbishing and load rail diagnostic testing.

Mathematical Modeling & Econometrics Sandbox

Interactive LaTeX formula renders and computational finance equations.

Heston Stochastic Volatility Model

Option Pricing

Coupled stochastic differential equations (SDEs) describing asset price evolution $S_t$ alongside mean-reverting stochastic variance $v_t$:

• $\mu$: Risk-free rate / Drift parameter

• $\kappa$: Speed of variance mean-reversion

• $\theta$: Long-term variance mean level

• $\sigma$: Volatility of volatility (Vol-of-Vol)

Cox-Ingersoll-Ross (CIR) Rate Model

Fixed Income

Short-rate SDE ensuring strictly non-negative interest rates via square-root volatility diffusion:

• Feller condition requirement: $2a b > \sigma^2$

• Guarantees strictly positive rate trajectories over long horizons.

Merton Jump-Diffusion Process

Fat-Tail Risk

Extends standard Black-Scholes geometric Brownian motion by introducing Poisson-driven discontinuous price jumps $dq_t$:

Used to capture discrete market crash events, earnings announcement shocks, and macroeconomic news announcements in options surface calibration.

Markov Decision Processes (MDPs)

Sequential Decisions

A framework for choosing actions over time when outcomes are uncertain. The optimal value of a state follows the Bellman optimality equation:

An MDP is defined by states, actions, transition probabilities, rewards, and a discount factor. Solving it with value iteration or policy iteration gives the best action to take in every state.

Engineering High-Performance Systems & Intelligent Infrastructure