Skills
Product Management
Product Discovery & Prioritization
Before proposing a solution, I assess it using a desirability, feasibility, and viability framework, then score it with RICE. At Hikaya this is usually where I start: sitting down with existing tools and workflows, figuring out what's actually broken, and ranking the fixes by impact rather than by whoever asked loudest. Total cost of ownership gets weighed in too, since the cheapest thing to build isn't always the easiest thing to maintain.
Roadmap & Stakeholder Alignment
A roadmap is only useful if everyone agrees on what matters, so most of my work happens before the roadmap exists: getting stakeholders to agree on a set of priorities. The job is the same whether it's a two-person team or a project with donors, country offices, and a global team all pulling in different directions. Once the priorities are set, Agile keeps us on track week to week: backlog grooming, two-week sprints, and retrospectives that shape what comes next.
User Research & Human-Centered Design
I try to talk to the people who'll use something before I write requirements for it: interviews, a few personas, and a journey map usually come first, before much goes into a Product Requirements Document. The grants management system I built at NRC started the same way, with workshops for country office staff and user stories drawn straight from what they said.
Data Architecture & Engineering
Data Modeling
I like to design ERDs and relational schemas early on, since a poorly designed data model tends to cost far more to fix later than to get right up front. For reporting, data marts following star and snowflake schemas are a good choice.
Metadata & Documentation
Metadata is data about data: the taxonomy and definitions that sit underneath a data dictionary. It's the same instinct behind HXL tags and IATI's codelists, where a column tagged #sector, or a value pulled from a shared code list, means the same thing no matter which organization published the dataset. The same problem often occurs getting different teams inside a single organization to agree on one standard dataset and actually use it, rather than each keeping its own slightly different version. The solution is a common data model, one shared schema that lets you combine data from systems, or teams, that were never set up to talk to each other.
API-First System Design
API-first design means agreeing on how other software talks to a system before building the system behind it. This helps set up the interface up front. That way, whoever builds against it next, a partner integration, another engineer has the guidance they need.
ETL & Data Integration
Most of my integration work is building ETL pipelines, the extract-transform-load jobs that move data between systems. I put validation and quality checks along the way, so a bad record gets caught early instead of quietly moving downstream.
Databases & Enterprise Systems
Databases & Spreadsheets
PostgreSQL and MySQL cover most of the structured data work. When the need is lighter, Airtable or MS Lists work well without setting up a fully fledged database, and Excel or Google Sheets are still the right call when a spreadsheet is genuinely all a project needs.
Enterprise Systems
I've configured and integrated MS Dynamics 365, Odoo, Salesforce, and Unit4 on different projects, usually inheriting the one an organization already had in place rather than setting up from scratch. Each has its own proprietary way of doing things, and most of the work is learning where those limits actually are.
M&E Platforms
I've worked hands-on with most of the major digital M&E platforms: ActivityInfo, DevResults, DHIS2, RedRose, and TolaData, which I helped build. Open source or proprietary changes the trade-offs, but the underlying problem stays the same: getting clean, structured data out of the field and into something that can be shared with a donor.
Analytics & Business Intelligence
Data Cleaning & Statistical Modeling
Cleaning and exploring a dataset usually takes longer than the modeling itself. That means fixing the missing values, the outliers, and the column that got renamed halfway through a project, then normalizing formats and units so the values compare. Which regression method to use depends on the question you're answering.
Cost-Benefit Analysis
When a management team is deciding between two solutions or approaches, whether that's build versus buy or which platform to standardize on, I like to perform a cost-benefit analysis. It puts the trade-offs on the table, so the call rests on the numbers rather than on whichever pitch sounded best in the room.
SQL Analysis
I write SQL with indexing in mind, since it makes a real difference once a table grows past a few thousand rows. A query that is indexed well can keep a dashboard loading in seconds rather than minutes.
Event & Log Analytics
I analyze event and log data, plus built-in or third-party analytics like Google Analytics and Search Console, to see how a product gets used and how people find it. Usually that helps surface the point where the journey breaks, a confusing step or a drop-off.
Reporting & Dashboard Design
I pick the simplest chart that answers the question being asked and use cross-tabulations by region, sex, or program type. On the dashboard itself, I limit each view to one clear metric, with consistent color coding and a limited number of charts. Filters help make it interactive, letting a country director and a field officer drill into the same dashboard instead of needing two.
Software Development & Cloud
Backend & APIs
Python handles most of the data-intensive work and the more involved backends, usually with Django and DRF. I define the REST APIs up front, so the frontend and any later integrations have something stable to build against. For smaller side projects I've started to work with full-stack TypeScript instead, with tRPC and Drizzle on Node.
Frontend
On the frontend I write TypeScript, mostly React with Next.js, and some Vue depending on what a codebase already uses. Either way I try to keep components reusable, so the same piece isn't rebuilt slightly differently across a project.
Delivery & Cloud
Git and CI/CD make up the delivery pipeline end to end. When a client is already built around Microsoft, PowerApps is usually the better call for low-code internal tools instead of a full custom build. I've deployed VMs across Azure, AWS, Google Cloud, and DigitalOcean depending on the project, with Entra ID handling identity on the Microsoft side.
Monitoring and Evaluation
Theory of Change & Results Chains
I've facilitated theory of change and results chain workshops with project and M&E teams: the exercise of mapping how day-to-day activities are supposed to add up to a long-term outcome. Most of the work is getting everyone to agree on what success looks like before anyone writes it down. The logical frameworks that come out hold up far better when every intervention is measurable from the start.
Quality Standards & Sampling
Where appropriate, I align quality assurance with OECD DAC standards, not as a box to tick but to make sure the data is built on a well-tested standard. I'm familiar with sampling strategies and understand how to accurately sample the target population using simple random, cluster, or stratified sampling.
Data Collection
ODK and XLSForm are the backbone of most of my offline data collection work, backed by ODK Central, Kobo, CommCare, or SurveyCTO depending on the deployment. I go with what an org already has, or the simplest setup its staff can realistically maintain.
Standards & Frameworks
IATI, FTS, and HXL structure how humanitarian data gets published, so it is usable by someone outside the organization that collected it. The Principles for Digital Development and the Digital Public Goods Standard go a layer deeper. They push me to avoid vendor lock-in, tools that only work for someone with a smartphone and reliable data, or a system that quietly falls apart once the funding that built it runs out.
GIS & Spatial Data
Mapping & Spatial Analysis
For deeper spatial analysis I work in QGIS or ArcGIS, the desktop GIS tools built for heavier geoprocessing and modeling. When a map just needs to ship inside a product, Mapbox and CARTO are the low-code option: quick to style and host, with OpenStreetMap data underneath. When a project needs to keep ownership of the whole stack and scale it without per-tile pricing, I build the map from open-source parts instead: MapLibre GL for the rendering and OpenFreeMap for the tiles. That's how the interactive maps on my trail pages are built.
AI & Machine Learning
Applied ML Foundations
My background is in statistics and that shapes how I think about machine learning. The same foundations, regression, probability, sampling, and what makes an estimate trustworthy, carry straight over into how models learn from data. It helps most not in building models from scratch but in reading them: what a model can realistically do, where its results hold up, and where the numbers are thinner than they look.
Hands-On with LLMs
Day to day this means prompting and evaluating models, fine-tuning small ones for a specific task, and using retrieval-augmented generation to check and measure against real data. I also build agent tooling with MCP, and run smaller models locally for lighter, everyday tasks where they're quick and cheap to run, and to keep a feel for how fast they're improving month to month.
Context Engineering
Most of the work that makes an agent useful isn't just the prompt, it's the context around it. That's the spec docs, skills, steering, and memory I write so a model understands a project the way a new teammate would. After each session I add what I learned back into that context, so the next run is more refined than the last. It builds up over a project: less of my time goes into steering every step, and more into reviewing work the agent runs.
AI Adoption & Governance
I help organizations adopt AI with intention rather than letting it spread in a rogue, shadow-IT state, guided by a staged maturity model I've written about. It runs from everyday chatbot fluency, to connecting AI to approved internal data and capturing what works as reusable skills a team can apply in context, to agentic processes that execute multi-step work under human oversight. Issues arise when orgs try to skip steps by reaching for agentic workflows before the data quality, governance, and staff trust are there to support them. Since an organization can only stretch so far past where its data and governance actually sit, I try to meet them where they are.
AI Usage Policy & Data Protection
Working with teams early on, I like to settle what shouldn't go into an AI tool: beneficiary PII, financial records, donor details, credentials. The rule of thumb I give non-technical staff is simple: if you wouldn't want the text turning up in a stranger's chat, it doesn't go in. Just as important is correcting the assumption that a paid plan protects your data automatically: on most consumer tools model training is on by default regardless of price, and the opt-out is on you. So a usable policy pairs clear data boundaries with the specific platform settings that actually enforce them.
IT Service Management
Service Design & Operations
I've worked on an IT operations team and across projects built around ITIL, which is where I got to put this into practice. I design services the ITIL-aligned way and run incident, problem, and change management on an ongoing basis, not only when something's already on fire. I use ITIL's own priority and severity standards to triage an issue and escalate it before it grows, then close the loop with a post-incident review so the root cause gets fixed instead of patched over and repeated.
Service Catalog & SLAs
I've built service catalogs and managed SLAs in both FreshService and ServiceNow. Getting the software set up is usually the easy part; getting an organization to agree on what an acceptable response time even is takes longer. Once that's settled, automation workflows do the actual triage, routing a request to the right queue and firing off an alert before an SLA is close to breaching. The analytics on response times and ticket volume are what point to where support needs to improve next.
Data Governance & Responsibility
Data Protection
I build data protection frameworks aligned with GDPR and the DPA, run DPIAs before a new system goes live rather than after, and practice data minimization by default.
Access & Architecture
Role-based access controls and secure-by-design architecture are critical first steps in anything that is built. Do No Harm goes further: before building anything, I also ask whether collecting this data at all could put someone at risk, not only whether the system storing it is secure.
Interested in bringing skills like these to a project? Get in touch