Published research
Open-access papers, free to read and cite. Most recent first.
19 papers
A low-cost radio interferometer for ionospheric monitoring near the equator
Equatorial ionospheric scintillation degrades satellite navigation and communication. We describe a low-cost, two-element radio interferometer built from software-defined radios to monitor scintillation over the Kenyan highlands. Six months of observations captured the post-sunset scintillation peak and its seasonal variation, agreeing with co-located GNSS scintillation indices. The design is reproducible for under 2,000 US dollars and suits distributed monitoring networks.
Cellulose nanofibre reinforcement of geopolymer cement from coal fly ash
Geopolymer cements offer a lower-carbon alternative to Portland cement but can be brittle. We reinforced a fly-ash geopolymer with cellulose nanofibres extracted from agricultural residue and measured flexural strength, toughness, and drying shrinkage. A one percent nanofibre loading increased flexural toughness by 34 percent and reduced shrinkage cracking without compromising compressive strength. We discuss dispersion methods that avoid nanofibre agglomeration.
Laser powder-bed fusion of a locally sourced titanium alloy for orthopaedic implants
Additive manufacturing of titanium implants can shorten supply chains if local feedstock meets quality requirements. We processed a South African Ti-6Al-4V powder by laser powder-bed fusion and characterised density, microstructure, and fatigue life against imported feedstock. After hot isostatic pressing, local material reached 99.7 percent density and comparable high-cycle fatigue strength. We report the process window and residual-stress mitigation strategy used.
Spatial clustering of schistosomiasis around irrigation schemes in the Awash basin
Irrigation expansion can increase transmission of schistosomiasis by extending snail habitat. Combining georeferenced infection data with canal maps in the Awash basin, we found strong spatial clustering of infection within 500 metres of perennial canals. Children living near canal junctions had 2.7 times the odds of infection. We recommend targeted snail control and water-contact reduction at identified hotspots.
Point-of-care loop-mediated isothermal amplification for malaria diagnosis in remote clinics
Microscopy and rapid diagnostic tests each miss a share of low-parasitaemia malaria infections. We evaluated a battery-powered loop-mediated isothermal amplification (LAMP) assay against nested PCR across five remote clinics. LAMP achieved 96.4 percent sensitivity and 98.1 percent specificity for Plasmodium falciparum, detecting infections missed by rapid tests in 11 percent of febrile patients. Turnaround was under 45 minutes without cold-chain reagents. We discuss operator training and quality assurance needed for routine use.
Explainable credit scoring from alternative data for thin-file borrowers
Thin-file borrowers lack the credit history that conventional scoring requires. We build a credit-scoring model from airtime top-ups, mobile-money flows, and handset metadata, pairing gradient-boosted trees with SHAP-based explanations so that adverse-action reasons can be disclosed. On a portfolio of 60,000 microloans the model improved default discrimination while producing human-readable reasons for each decision. We discuss fairness auditing across gender and region.
Agent networks and the last mile of digital financial inclusion in Nigeria
Mobile-money agents are the physical interface of digital finance, yet agent liquidity and reliability vary widely. Drawing on transaction logs from 1,200 agents and a household survey in three Nigerian states, we show that agent density explains more of the variation in active-account rates than smartphone ownership. Cash-out failures, driven by agent illiquidity, were the leading reason customers abandoned digital wallets. We propose liquidity-rebalancing incentives and measure their simulated effect on service reliability.
Teaching numerical methods with open, executable notebooks
Executable notebooks let students see numerical methods run, edit parameters, and observe convergence directly. We report a semester of teaching introductory numerical methods with a set of open notebooks covering root finding, interpolation, and ordinary differential equations. Compared with a prior cohort taught with static notes, students using notebooks scored higher on applied problems and reported greater confidence implementing methods. We share the notebook set and describe the assessment design.
Design rules for reliable low-cost IoT PCBs
Low-cost IoT hardware often fails for avoidable layout reasons: poor decoupling, weak antenna keep-outs, and thermal relief mistakes on ground pours. Drawing on a review of 30 community hardware projects, we distil a compact set of PCB design rules that improve reliability without raising fabrication cost, and we provide an open checklist and reference layout. We quantify the effect of each rule on a test board through radiated-emissions and thermal measurements.
Community health worker triage and time to care for childhood pneumonia in western Kenya
Delays in seeking care worsen outcomes for childhood pneumonia. We evaluated a community health worker programme that used a symptom-based triage aid and referral vouchers across 60 villages. Median time from symptom onset to appropriate care fell from 52 to 29 hours, and referral completion rose from 48 to 74 percent. We assess the programme's fidelity and the conditions under which the triage aid changed caregiver behaviour.
A reproducible pipeline for interpolating sparse rainfall data for smallholder advisories
Weather advisories for smallholder farmers depend on rainfall estimates in areas with sparse gauge coverage. We present an open, reproducible pipeline that blends gauge data with satellite rainfall estimates using kriging with external drift, and we evaluate it against withheld gauges across two seasons in western Kenya. The blended product reduced mean absolute error by 27 percent over satellite-only estimates and produced calibrated uncertainty bands suitable for advisory thresholds. Code and data-preparation steps are shared so the workflow can be rerun in new regions.
A low-power LoRaWAN sensor node for off-grid environmental monitoring
Remote environmental monitoring needs sensor nodes that run for months on small batteries and report over long distances. We describe an open-hardware LoRaWAN node combining a low-power microcontroller, calibrated temperature, humidity, and soil-moisture sensors, and an energy budget dominated by deep-sleep current. Measured average consumption of 38 microamps yields an estimated 14-month lifetime on a single 3.6-volt cell at a fifteen-minute reporting interval. We share the schematic, firmware, and calibration procedure, and we report field results from a three-month deployment on a smallholder farm.
Community-managed sand dams and dry-season water security in semi-arid Kitui
Sand dams store water within riverbed sediment, offering a low-cost source through the dry season. Surveying 28 community-managed sand dams in Kitui, Kenya, we found that households near functioning dams walked 46 percent less far for water in the dry season and reported fewer days without any household water. Governance quality, not dam age, was the strongest predictor of sustained function. We identify maintenance-committee practices associated with long-term performance.
Farmer-managed natural regeneration and household resilience in the Nigerien Sahel
Farmer-managed natural regeneration (FMNR) has re-greened large areas of the Sahel, but its household-level welfare effects are debated. Using a panel of 900 households across Niger, we estimate the effect of FMNR adoption on food security and income diversification. Adopting households experienced 22 percent fewer months of food shortfall and greater income from tree products, with effects strongest in the driest sites. We discuss tenure conditions that sustain adoption.
A modular Python framework for real-time simulation of mechatronic systems
Teaching and prototyping mechatronic systems benefits from simulations that run in real time against real hardware. We present an open-source Python framework that composes mechanical, electrical, and control subsystems as interchangeable modules with a fixed-step solver and a hardware-in-the-loop bridge. On a two-wheeled balancing robot, the framework sustained a one-kilohertz control loop on a single-board computer while streaming state to a browser dashboard. We describe the module interface, the solver design, and the latency budget that keeps simulation and hardware in step, and we release worked examples so readers can reproduce each result.
Grid-forming inverters for weak distribution networks: a field study in the East African interconnector
As inverter-based renewable generation grows on weak distribution feeders, conventional grid-following control can struggle to maintain stability during faults and rapid load changes. We report a six-month field trial of grid-forming inverter control on a 2 MW solar plant connected to a low-short-circuit-ratio feeder. The grid-forming configuration improved voltage recovery after faults and provided synthetic inertia that reduced frequency excursions during islanding events. We present measured fault-ride-through behaviour, compare it against grid-following operation on the same site, and outline protection-coordination changes required for wider deployment.
Techno-economic assessment of hybrid solar-diesel mini-grids for rural health facilities in Kenya
Reliable electricity is a precondition for functional rural health facilities, yet many remain dependent on intermittent grid supply or costly diesel generation. This study evaluates hybrid solar photovoltaic and diesel mini-grids for 42 dispensaries across three counties in central Kenya. Using measured load profiles and one year of on-site irradiance data, we size photovoltaic arrays, battery storage, and diesel backup, then compare levelised cost of electricity across configurations. A solar fraction of 0.7 with two hours of battery autonomy minimised lifecycle cost while keeping unmet load below 2 percent. Compared with diesel-only baselines, the optimal hybrid reduced fuel consumption by 61 percent and lifecycle emissions by 54 percent. We discuss financing structures and maintenance models that make such systems viable under county health budgets.
Satellite-based crop-yield nowcasting for smallholder maize in the Rift Valley
We describe a machine-learning pipeline that combines Sentinel-2 imagery, rainfall estimates, and soil data to nowcast smallholder maize yields at sub-county resolution. Trained on three seasons of ground-truth harvest measurements, the model explained 71 percent of yield variance and flagged drought-stressed zones four to six weeks before harvest. We discuss how such early signals can inform anticipatory action by county agriculture offices.
Low-resource speech recognition for Kiswahili clinical dictation
Clinical documentation in East African hospitals is time-consuming and often incomplete. We present a speech-recognition system for Kiswahili clinical dictation trained with only 40 hours of transcribed audio by combining self-supervised pretraining with targeted data augmentation. On a held-out set of ward-round recordings, the system reached a word error rate of 14.2 percent, a relative improvement of 38 percent over a supervised baseline. We analyse error patterns around code-switching between Kiswahili and English medical terms and describe a lightweight deployment that runs on commodity hardware without cloud connectivity.