Roseline – .NET, C#, ASP.NET Web API

An outbound integration service that connects internal services with external telecom providers. The service processes data bundle purchase commands, communicates with provider APIs, tracks transaction status, and publishes transaction outcomes back to internal services. Built around Domain-Driven Design and event-driven architecture, it provides resilient retry and exponential backoff mechanisms for transient failures and progressively queries pending transactions until a final success or failure state is reached.

• Designed the service architecture and documented the solution using Miro and Notion.
• Developed the integration using C#/.NET Worker Service and event-driven messaging.
• Integrated with Safaricom APIs for data bundle purchases and transaction-status queries.
• Developed additional retry handling for known recoverable non-transient failures.
• Implemented progressive polling for pending transactions, with escalation when a transaction could not be resolved.
• Created user stories and technical documentation and reviewed the architecture with the engineering team before implementation.
• Implemented idempotent purchase processing using Command ID, preventing duplicate data bundle purchases during retries and repeated event delivery.
• Implemented retry and exponential backoff for transient failures and additional retry handling for recoverable non-transient failures.

An AI software engineering platform focused on training and evaluating large language models for code generation and real-world software development tasks. The product supports model improvement across reasoning, coding accuracy, debugging, full-stack application generation, and software behavior through structured feedback, benchmark-based evaluation, and iterative correction.

• Trained and aligned AI models using RLHF/RLFHD-style human feedback to improve reasoning and code-generation quality.
• Evaluated AI-generated code and full-stack applications for correctness, API integration, UI/UX fidelity, performance, reliability, scalability, and maintainability.
• Performed SWE-bench-style evaluations, testing edge cases and real-world repository behavior.
• Used AI coding agents such as Cline and Claude Code to work with large repositories and iteratively improve generated solutions.
• Designed and refined prompts to improve model reasoning, debugging, and code-generation accuracy.
• Built, curated, and annotated code datasets across languages including Python, JavaScript, Java, C#, C++, C, Go, and Rust.
• Reviewed model outputs, identified failure cases, and provided structured feedback, annotations, and correction loops to improve future model behavior.
• Helped models better understand real-world engineering workflows involving REST APIs, asynchronous processing, testing, GitHub, and repository-based development.

A data bundle fulfillment platform that serves as the central source of truth for bundle disbursement across multiple purchase and benefit models, including cash-based purchases and recurring customer benefits. The product manages the full fulfillment lifecycle and maintains the current state of every fulfillment request using an event-sourcing architecture, with domain events persisted in Cosmos DB and Change Feed used to react to state changes and publish updates to other services. Azure Application Insights provides observability through custom events and distributed tracing, enabling monitoring and troubleshooting across the fulfillment process.

• Designed and developed backend functionality for the Data Bundle Fulfillment Service as the authoritative source for bundle disbursement state.
• Implemented event-sourced domain models where data bundle fulfillment and disbursement information are represented as aggregates and updated through commands and domain events.
• Processed fulfillment lifecycle events, including completed fulfillment events, and persisted aggregate state in Azure Cosmos DB.
• Used Cosmos DB Change Feed to detect state changes and publish corresponding events to downstream services.
• Implemented asynchronous communication using Azure Service Bus queues and topics for commands, events, and cross-service messaging.
• Developed event mappers to transform internal domain events into integration contracts consumed by other M-KOPA services.
• Configured fulfillment queues and subscriptions used by upstream services to submit data bundle fulfillment requests.
• Used Bicep Infrastructure as Code to provision and configure Service Bus queues, topics, subscriptions, event registrations, and supporting Azure infrastructure.
• Implemented observability using Azure Application Insights, including custom events for tracking fulfillment states and processing stages.
• Added distributed logging and telemetry to trace fulfillment requests throughout the complete disbursement workflow.
• Worked with Azure DevOps for source control, work tracking, pull requests, CI/CD pipelines, and sprint delivery.
• Contributed to the reliability and compatibility of the service by supporting existing integrations while evolving the event-driven fulfillment architecture.

A backend-for-frontend platform that serves as the integration layer between a mobile application and downstream microservices supporting data bundle purchases. The product orchestrates the customer journey across cash, loan-based, and benefit bundles by coordinating eligibility checks, customer identity, loan applications, offer generation, and fulfillment. Using an event-driven architecture, it processes commands and events to coordinate downstream services, track fulfillment outcomes, and return the final purchase status to the mobile application.

• Developed the BFF APIs used by the mobile application for the end-to-end data bundle purchase journey.
• Integrated with the Eligibility Service to determine whether customers qualify for cash, loan, or benefit data bundles.
• Integrated with the Identity Service to retrieve customer details such as phone number and account information.
• Orchestrated the loan application and data bundle offer flow using commands and domain events.
• Resolved product information and available bundle offers before presenting them to customers in the mobile application.
• Published customer bundle selections to the Data Bundle Fulfillment Service and consumed fulfillment success/failure events.
• Coordinated communication across multiple downstream microservices while keeping the mobile application isolated from internal service complexity.
• Implemented reliable event and command handling for distributed service communication.

An internal compliance automation platform for validating Windows builds before release. The product allows product teams to submit build information, run automated scans, review compliance findings through an Angular dashboard, download reports, and create or assign Azure DevOps bugs and incidents to the appropriate developers. The platform helps identify policy and compliance issues across binaries, code, metadata, language usage, and other release requirements before builds are shipped.

• Built and maintained backend API services for build management, product management, user management, scan operations, and incident/bug management.
• Developed Angular and Angular Material components for scan dashboards, recent scans, paginated scan results, individual scan details, bug creation, and developer assignment.
• Developed Azure Functions and Durable Functions to process scan results and support scan orchestration.
• Integrated Azure Blob Storage for scan output processing and Azure Service Bus for event and command communication between the APISCAN backend and the scanning pipeline.
• Integrated Azure DevOps for bug and incident creation and assignment.
• Added application logging and monitoring using Azure Application Insights and worked on both backend and frontend features end to end.
• Helped modernize a legacy VB macro and Excel-based Windows compliance scanning process into a full-stack automated platform. The solution automated scan execution, result processing, dashboard visualization, bug/incident workflows, and CSV/Excel report generation, significantly reducing the manual effort required by Product Managers. The automated scan itself was reduced from about 15 hours to 5 hours, while the broader workflow that could previously take days was made substantially faster and easier to manage.