Kelvin Mihingo N.

Software Engineer building AI-agent platforms and the frontend systems behind them.

I own and ship ambiguous, cross-team engineering initiatives from architecture through production adoption and general availability.

Selected work

Products and platforms I have built, owned or led — each with its context, my contribution and where it stands today.

Plate 01Founder / Software Engineer

Police Service Management System

BuildingPublic reporting portal live

The live PSMS public reporting portal on desktop: a navy hero reading “Report, Track, Prevent” above search filters for crime type, county and category.
The same portal on a phone: the “Report, Track, Prevent” headline above the stacked crime search filters.
Fig. 1.1 — kenya.psms.africaThe live public reporting portal, captured October 2026. Portal statistics are cropped out.

A full-stack public-safety platform for police services, bringing occurrence reporting, investigations, case management, assignments, analytics and command dashboards into one system.

Role
Founder / Software Engineer
Where it stands
In development. The public reporting portal is live at kenya.psms.africa.

What I did

  • AI-assisted workflows for case analysis, grounded summaries, report intelligence and decision support.
  • Production-oriented foundation: automated testing, secure connectivity, notifications and a Turborepo/Yarn monorepo.
  • Next.js and Relay over a .NET 8 / HotChocolate GraphQL API, with Entity Framework Core and PostgreSQL.
  • Azure Front Door Premium, WAF and private connectivity, with Keycloak for identity.

Built with Next.js · TypeScript · Relay · GraphQL · .NET 8 · HotChocolate · Entity Framework Core · PostgreSQL · Docker · Azure · Keycloak

Plate 02Microsoft · 2021 – present

Shared reasoning experience for Microsoft 365 Copilot agents

ShippedArchitecture through GA

Fig. 2.1 — Scope of the shared reasoning experience
Scope of the shared reasoning experienceA conceptual view of the scope described above — one shared reasoning experience serving first- and third-party agents across four Microsoft 365 surfaces. Not an internal architecture diagram.First-party agentsThird-party agentsShared reasoning experienceReusable components & APIsTeamsOutlookWordPowerPoint
    • First-party agents
    • Third-party agents
    • Shared reasoning experienceReusable components & APIs
    • Teams
    • Outlook
    • Word
    • PowerPoint

A conceptual view of the scope described above — one shared reasoning experience serving first- and third-party agents across four Microsoft 365 surfaces. Not an internal architecture diagram.

Read the diagram as text
  • First-party agents → Shared reasoning experience
  • Third-party agents → Shared reasoning experience
  • Shared reasoning experience → Teams
  • Shared reasoning experience → Outlook
  • Shared reasoning experience → Word
  • Shared reasoning experience → PowerPoint
Fig. 2.2 — Also at Microsoft — load time of a key Microsoft 365 app
Before
about 4 s
After
under 1.5 s

Over 60% faster, measured with the Browser Performance API.

Standardising how LLM reasoning is presented across Microsoft 365 Copilot agents — for first- and third-party agents, in Teams, Outlook, Word and PowerPoint.

Role
Initiated and led the engineering effort. Primary engineering owner, later co-lead as the effort scaled.
Where it stands
Taken from early architecture through general availability.

What I did

  • Scaled the shared reasoning experience across Teams, Outlook, Word and PowerPoint, for first- and third-party agents.
  • Built reusable platform frameworks and developer abstractions: plug-and-play APIs, reusable components, agentic workflow abstractions.
  • Authored specs and cross-team architecture guidance: service-client contracts, message consistency, rollout dependencies, production blockers.
  • Designed product telemetry (engagement, latency, dwell time, step interactions) that informed staged rollout and GA decisions.
  • Established Playwright-based end-to-end regression coverage.

Built with React · TypeScript · Fluent UI v9 · Zustand · GraphQL · Playwright · Storybook · KQL / Azure Data Explorer

Plate 03Minsait (formerly Indra Sistemas) · 2019 – 2020

Revenue, payments and prepayment systems for Kenya Power

Delivered

legacy debt migrated to a microservices architecture
USD 5M+
countries using the prepayment-retry microservice
3
Kenya · Nigeria · Zimbabwe

Backend engineering for Kenya Power's revenue collection and customer systems — migrating legacy debt, integrating payments and keeping prepayment requests resilient.

Role
Backend Engineer for client Kenya Power & Lighting Company (KPLC). Led the Interfaces Module (Indra InCMS to third-party systems).

What I did

  • Built a middleware application to migrate USD 5M+ of legacy debt to a microservices architecture.
  • Built a resilient microservice for failed prepayment API requests (Itron), used in Kenya, Nigeria and Zimbabwe.
  • Integrated card-present payment capabilities into the revenue collection system.
  • Redesigned the customer onboarding and self-service flow for tenancy accounts.

Built with Java · Spring Boot · microservices · SOAP/SAP integration · Oracle · Jenkins

How I work

My work spans Microsoft 365 product and service-integration engineering, reusable AI-agent architecture, and cross-product platform capabilities, with a focus on scalable frontend systems and LLM-powered experiences.

  1. Own the whole arc

    I take ambiguous, cross-team initiatives from first architecture through production adoption and general availability.

    Evidence Initiated the effort to standardise LLM reasoning UX across Microsoft 365 Copilot agents — primary engineering owner, later co-lead as it scaled.

  2. Build the platform, not the one-off

    I favour reusable frameworks, components and APIs that other teams can build on.

    Evidence Plug-and-play APIs, reusable components and agentic workflow abstractions for the Microsoft 365 AI-agent ecosystem.

  3. Measure, then decide

    I instrument what I ship, so rollout and performance decisions rest on evidence.

    Evidence Telemetry on engagement, latency, dwell time and step interactions informed staged rollout and GA; Browser Performance API measurement behind a 60%+ faster load.

  4. Align in writing

    I put contracts, dependencies and risks on paper early, where every team can see them.

    Evidence Authored specs and guidance on service-client contracts, message consistency, rollout dependencies and production blockers.

  5. Grow the people around me

    I invest in the engineers and educators who come next.

    Evidence Mentored five engineering interns; led AI and software-engineering upskilling for Kenyan university lecturers; contributed to curriculum reviews.

Capabilities

Grouped by the problems they solve. Each is backed by the work above.

  • AI-agent & LLM experiences

    Standardised LLM reasoning UX across Microsoft 365 Copilot agents; built LLM experiences over Word, PowerPoint, meetings, tasks and Teams messages; AI-assisted case analysis in PSMS.

    AI/LLM engineering · AI-agent experiences · LLM reasoning UX

  • Frontend & platform architecture

    Reusable frameworks, plug-and-play APIs and components for the Microsoft 365 AI-agent ecosystem; a key Microsoft 365 app made over 60% faster.

    React · TypeScript · JavaScript · Fluent UI React v9 · Zustand · Storybook · frontend architecture · Browser Performance API

  • Full-stack product engineering

    PSMS end to end — Next.js and Relay over a .NET 8 / HotChocolate GraphQL API on PostgreSQL.

    Next.js · Relay · GraphQL · HotChocolate · C# (.NET) · Node.js · PostgreSQL · SQL & NoSQL · full-stack development

  • Integration & backend systems

    Middleware migrating USD 5M+ of legacy debt; card-present payments; a resilient service for failed Itron prepayment requests.

    Java · Spring · Maven

  • Delivery & quality

    Telemetry-informed staged rollout and GA decisions, Playwright end-to-end regression coverage and production support.

    Playwright · Azure DevOps · CI/CD · telemetry · KQL / Azure Data Explorer · progressive delivery · incident response

  • Leadership

    Led architecture, design and code reviews and aligned contracts across teams; mentored five engineering interns.

    technical leadership · cross-team collaboration · mentoring

Track record

The roles, the results and the people along the way.

In numbers

  • 60%+faster loadAbout 4s to under 1.5s on a key Microsoft 365 app, measured with the Browser Performance API.
  • 4Microsoft 365 surfacesTeams, Outlook, Word and PowerPoint.
  • 5engineering interns mentoredSupporting the next generation of engineers.
  • $5M+in legacy debt migratedUSD 5M+ migrated to microservices for KPLC.
  • 3countries servedPrepayment-retry microservice used in Kenya, Nigeria and Zimbabwe.

Experience

  1. Microsoft

    Software EngineerJan 2021 — Present5 yrs 9 mosNairobi · HybridBuilding shared AI-agent experiences and the platform foundations behind Microsoft 365 Copilot.Role details
    • Initiated and led the engineering effort to standardize LLM reasoning UX across Microsoft 365 Copilot agents, from early architecture through GA. Primary engineering owner, later co-lead as the effort scaled.
    • Scaled the shared reasoning experience across Teams, Outlook, Word and PowerPoint, for first- and third-party agents.
    • Built reusable platform frameworks and developer abstractions for the Microsoft 365 AI-agent ecosystem: plug-and-play APIs, reusable components, agentic workflow abstractions.
    • Improved performance of a key Microsoft 365 app by over 60% (about 4s to under 1.5s) using the Browser Performance API.
    • Authored technical specs and cross-team architecture guidance: service-client contracts, message consistency, rollout dependencies, production blockers.
    • Designed and implemented product telemetry (engagement, latency, dwell time, step interactions) that informed staged rollout and GA decisions.
    • Led architecture, design and code reviews; aligned contracts across teams; planned rollouts; supported production. Established Playwright-based end-to-end regression coverage.
    • Built AI-powered experiences integrating LLMs with Microsoft 365 content and productivity data (Word, PowerPoint, meetings, tasks, Teams messages).

    Built with React · TypeScript · Fluent UI v9 · Zustand · GraphQL · Playwright · Storybook · Azure DevOps · KQL / Azure Data Explorer · LLM reasoning UX

  2. Minsait (formerly Indra Sistemas)

    Backend EngineerJan 2019 — Dec 20202 yrsNairobiClient: Kenya Power & Lighting Company (KPLC)Microservices, payment integrations and resilient utility systems for Kenya Power.Role details
    • Built a middleware application to migrate USD 5M+ of legacy debt to a microservices architecture.
    • Integrated card-present payment capabilities into the revenue collection system.
    • Redesigned the customer onboarding and self-service flow for tenancy accounts.
    • Led the Interfaces Module (Indra InCMS to third-party systems); built a resilient microservice for failed prepayment API requests (Itron), used in Kenya, Nigeria and Zimbabwe.

    Built with Java · Spring Boot · microservices · SOAP/SAP integration · Oracle · Jenkins · payments

Beyond the code

  • Engineering mentorship

    Mentored five engineering interns.

  • Teaching the teachers

    Led AI/software-engineering upskilling for Kenyan university lecturers.

  • Shaping the curriculum

    Contributed to the CS curriculum review at Baraton University and to the Kenya CBC Computer Studies curriculum via KICD.

Education

BSc Computer Science

University of Nairobi

2014 — 2018 · Second Class Upper

Let’s talk.

About a role, a product or a hard problem worth building. Email is the fastest way to reach me.

kmihingo22@gmail.com

GitHub profile linked; repositories are private.

Based in
Nairobi, Kenya
Open to relocation
Netherlands · Germany · UAE · UK · Ireland · Australia