Research data infrastructure, end to end.

Start4All Phase 1 — Cameroon Country Data Management

End-to-end Cameroon data operations for Start4All Phase 1 — a 7-country TB intervention programme evaluating near-POC molecular diagnostics and sputum pooling. Evidence fed the WHO 9 March 2026 recommendations; LSTM's 27 February 2026 commentary explicitly cited Start4All.

• Data workflows across screening and treatment event forms with dynamic data-dictionary-driven recoding (no hard-coded value→label maps).
• Calculated delay metrics: screening → diagnostic → treatment initiation timelines.
• Multi-level TB outcome classification (cure / failure / lost-to-follow-up) with competing-event logic over 6-month follow-up.
• Prioritisation flagging surfacing high-value data-entry errors for enumerator QA.

TB Reach Wave 10 Cameroon — Data Management

Cameroon TB Reach Wave 10 data backbone — community-based active case finding using chest X-ray AI triage paired with sputum pooling for molecular confirmation.

TB Reach Wave 11 — Health Camps & Facility Screening

Largest and most operationally complex BCHPR data system — 19,673 lines integrating 5 REDCap projects across community health camps, primary health facility screening, and prison screening in 7 regions of Cameroon. Target: 494,400 screenings and 105,646 diagnostic results.

• 5 integrated REDCap projects (Health Camps · Location · Primary Health Screening · Primary Health Results · Daily Attendance).
• Dual QR-code schemes — screening QR for participant ID and testing QR for specimen ID — enable flexible post-collection linkage.
• Spirometry PDF extractor (1,743 lines) using pdfplumber with French / English / hybrid detection, SHA-256 deduplication, and SQLite file-hash index.
• Windows COM Outlook email processor that auto-ingests monthly site reports from regional coordinators with MMM-YYYY regex.

GHIT FujiLAM II — Multi-Country Diagnostic Accuracy Study

21,036-line data platform for a diagnostic accuracy study evaluating FujiLAM, Alere LAM, GeneXpert, and molecular tests across 12 sites in Cameroon and Nigeria — supporting WHO prequalification evidence for next-generation TB diagnostics.

NPOC — Near-Point-of-Care TB Testing Pilot

12,613-line system running the Near Point of Care TB testing pilot at community health centres — evaluating LC-NAAT alternatives (Pluslife MiniDock, Molbio MTB Ultima, PortNat) against microscopy and centralised GeneXpert, with co-integrated spirometry screening for TB / asthma / COPD.

Pluslife MiniDock & Molbio MTB Ultima — Near-POC Evaluation

2,299-line pipeline (plus 1,039 LOC in the parallel retrospective-pooled track — 3,338 LOC across the full Pluslife programme) integrating Pluslife MiniDock and Molbio MTB Ultima CSV exports into unified diagnostic datasets for pooled and individual workflows — the data backbone of the 1,097-participant Clinical Infectious Diseases 2025 paper.

Image Quality Study — X-ray Machine Standardisation

5,173-line multi-country study comparing digital X-ray machine image quality (PRORAD vs Min Xray vs Fuji X-air) with DICOM metadata extraction and AI score differencing — informs equipment procurement and WHO guidance on digital X-ray adoption.

HIV Viral Load Pooling — Abbott m2000 Pipeline

4,785-line pipeline (the 3,595-line VIRAL LOAD DATA CLEANING.py is the largest single file in the project) ingesting Abbott m2000 HIV-1 viral-load results through REDCap — powers the 12,396-specimen pooled-testing study published in Scientific Reports 2025 (+80% testing capacity, 0.01% misclassification).

GeneXpert MTB/RIF — 16-Lab Pooled Testing Pipeline

4,942-line pipeline ingesting multi-site GeneXpert MTB/RIF CSV exports, harmonising French / English bilingual records, and consolidating pooled + individual TB test results — the data backbone of the 2026 openRxiv preprint (first & corresponding author).

Truenat Ultra — Rapid TB Diagnostics Pipeline

1,828-line pipeline for Truenat Ultra rapid TB PCR instruments — incremental ingestion with MD5-based change detection, SQLite state tracking, and Polars-optimised consolidation across a distributed lab network.

Global Outcome Tracking Hub — 7-Project TB Consolidation

Mission-critical ETL hub aggregating TB patient-outcome data from 7 independent research projects into a unified REDCap database with longitudinal patient-journey tracking — 116,000+ records across 390+ treatment sites.

Specimen Transport & Chain-of-Custody

2,328-line specimen transport and chain-of-custody tracking system linking sending and receiving REDCap records with study-ID validation and transport-quality assurance — enables 100% specimen accountability across the CHPR lab network.

Wave 11 Health Camps — Community TB Screening Cleaner

3,861-line data-cleaning and importation pipeline for the TB Reach Wave 11 community health camps programme — supports the 494,400-screening target across 8 regions of Cameroon with enumerator-level performance tracking.

Chest X-ray AI + Pooled Molecular Testing (Union 2025)

Demonstrated that combining AI-scored digital chest X-ray triage with sputum pooling extends molecular TB diagnostics into resource-limited community settings — with materially lower per-case cost.

QR-Code Patient & Specimen Tracking System

Operational innovation deployed across 365+ primary health facilities and 25+ GeneXpert laboratories — now under active consideration by Cameroon's National TB Programme as national best practice.

Request Management System — Power Platform End-to-End

Production-grade Microsoft Power Platform solution running the full lifecycle of financial and operational requests at BCHPR — submission, document validation, executive approval, payment, issue tracking, and reporting — built on SharePoint Lists, Power Automate, Teams Approval Cards, and a Power BI semantic model.

Inventory & Asset Management — Manager.io + M365

Manager.io-integrated inventory and asset-management system for CHPR logistics across multiple Cameroon sites — with Microsoft 365 ecosystem orchestration (Lists, Power Automate, Power BI) and automated procurement workflows.

Power BI Dashboards — 1,953 DAX measures across 8 models

Production Power BI estate with 8 major semantic models totalling 147 tables and 1,953 DAX measures — underpinning clinical decision-making, operational tracking, and executive reporting across every BCHPR research programme and the Admin Team Request Management System.

28 Apache Airflow Production Pipelines

Production Airflow 3 orchestration for every BCHPR data pipeline — 28 DAGs running from every 10 minutes to 3-hourly, with a custom FileMtimeSensor, Slack + email alerting, and a reusable DAG factory pattern.

• 28 production DAGs covering Wave 11, GHIT FujiLAM II, NPOC, Start4All, Viral Load, Xpert, Truenat, Pluslife, Image Quality, Inventory Management, Specimen Transport, TB Treatment, Chest X-ray, Culture, TBRL Lab, Collaborators, UCD DDE Check, Global Outcome, and maintenance jobs.
• Custom `FileMtimeSensor` — rescheduling sensor that triggers on file / directory mtime changes, with SHA-based baselines stored in Airflow Variables (no false positives on first run).
• `build_script_dag` factory pattern — parameterised DAG construction keeps 1,242 lines maintainable across 28 DAGs.
• Cadences — every 10 min (Culture) · every 15 min (cleanup, UCD DDE) · 30 min (user activity, specimen transport, lab PDF) · 45 min (FujiLAM UA) · hourly (most pipelines) · 3-hourly (reports) · daily (DQ scoring, collaborators).

data_quality_manager.py — Data-Quality Engine

REDCap validates one field at a time, at entry. It cannot express “this specimen date precedes the enrolment date on a different form”, or “these two record IDs are the same participant”. This 14,274-line engine does, and it remembers — every issue it raises has an open date, a close date, and a resolution time.

• Seven-dimension quality taxonomy (completeness, validity, accuracy, consistency, timeliness, uniqueness, integrity) with 35 error categories mapped onto it.
• Fluent QueryBuilder over a declarative, immutable schema — referencing an undeclared field fails at build time rather than silently yielding null.
• Six selectable scoring algorithms, defaulting to defect density (failures per opportunity to fail) rather than ticket-closure velocity.
• SQLite tuned for SharePoint-synced drives: WAL where safe, DELETE journaling where it is not, busy timeouts, and configurable integrity-check depth — a full check on a multi-GB tracker can stall for tens of minutes.

MANAGER.py — Manager.io Accounting API Client

15,661-line Manager.io API client. Full create / update / delete across 18 document types — sales quotes, orders, invoices, credit notes, delivery notes and receipts; purchase quotes, orders, invoices and debit notes; inventory items, locations, transfers, write-offs, goods receipts, production orders and starting balances; and payments. Customers, suppliers, projects, accounts, assets and tax codes are exposed read-only.

data_utilities.py — Shared DataFrame Toolbox

The dependency root of the library — 9,746 lines that import nothing else in the codebase and that everything else imports. 84 public callables covering REDCap activity-log parsing, data-dictionary recoding, checkbox column expansion, export diffing, and file I/O that survives a SharePoint sync.

redcap_api.py — Typed REDCap API Facade

REDCap exposes one POST endpoint per project, dispatched by form parameters rather than typed routes, with no retry, no pagination, and no bulk file endpoint. This 6,561-line facade wraps 16 of those endpoints behind a typed surface of 42 exported names.

dump_upload_pipeline.py — Mobile-Dump Ingestion

Built to end a specific failure: the notebook it replaced created duplicate patients three ways. Repeated QR codes were suffixed −1, −2 and uploaded as separate people; a field phone syncing straight to REDCap produced a second unlinked record for someone already dumped; and merging forms before upload multiplied rows. Duplicate enrolment inflates the denominator, which is an endpoint-validity problem, not a tidiness one.

m365_graph.py — Microsoft Graph Client

5,573-line Microsoft Graph client wrapping 73 distinct operations across Entra ID users and groups, SharePoint sites and lists, Microsoft Planner, and Teams. Planner is the centre of gravity at 43% of the file — activity tracking across study sites.

date_utils.py — Date Engine and Power BI Calendar

5,220 lines handling the dates that actually arrive: French month names from Francophone Cameroon, Excel serials, paper forms re-entered at the wrong precision, and REDCap fields whose _dmy suffix controls display only — the API returns ISO regardless, so a hand-exported Excel file parses differently from an API pull.

study_id_patterns.py — Study-ID Pattern Registry

The most-imported module in the codebase — 35 files depend on it. 27 registered patterns (16 participant and specimen ID forms, 11 site codes) covering identifiers that arrive from QR scanners, free-text notes and typed entry, where a transposed digit produces something structurally valid and completely wrong.

paths_manager.py — Cross-Platform Path Resolution

Every project script reads and writes under a shared SharePoint-synced tree. Hard-coding those paths breaks the moment a script runs on another machine, under WSL, or inside Airflow. 586 lines resolve 213 named paths across 27 project groups, from one environment variable.

api_manager.py — Credential Registry

482 lines holding 105 credential attributes across 16 project groups, read from the environment and never from source. Gives every token a name and a group so an environment can be audited before a pipeline fails halfway through a run.

ICRC — Protection Data & Analytics

Managed protection data systems supporting humanitarian operations during Cameroon's Anglophone Crisis — under ICRC's rigorous data protection standards, coordinating with armed groups, military, and civilian authorities.

HIV-Free — Budget & M&E (CDC/PEPFAR)

Budget monitoring and M&E tooling for the HIV-Free project across 20+ health facilities in Northwest Cameroon — built the indicator tracking and variance-analysis tools used across the programme.

ReAnalytics Consulting — Independent Advisory Practice

Independent practice positioned as Data Analytics and Computerized Accounting consultancy. Predates BCHPR by six years; runs in parallel today. Holds the QuickBooks Online Certified ProAdvisor designation as a firm-level credential (issued 15 March 2025; exam passed by Ngha as principal practitioner). The operational proof of the accounting → data hybrid.

• Service lines: QuickBooks + Manager.io implementation and training, Excel consultancy and training, data analysis and consultancy, data systems design, Power BI consultancy and development.
• Active engagements span NGOs and SMEs across QuickBooks Online and Manager.io — work is kept private at the request of clients; details available on request.
• Firm-held QuickBooks Online Certified ProAdvisor designation (Intuit, 15 March 2025) — the professional advisor tier reserved for accountants serving clients, distinct from the basic Certified User track.
• End-to-end engagement model: chart-of-accounts design, migration from spreadsheets to cloud accounting, bank-feed automation, monthly close, management reporting, and BI dashboards on top of the ledger.