# Mati Data

> I help teams design, build and govern data platforms they can trust. Mati Data is the data practice of Matiwos Desalegn (Mati), a data architect and senior data engineer in Addis Ababa, Ethiopia.

Senior Data Engineer at Kifiya Financial Technology in Addis Ababa, running lakehouse, CDC and warehouse platforms for digital lending since 2024. I take data work from architecture and governance to production pipelines and analytics. AI engineering is my second discipline.

5+ years building with data and software (since 2021), including production data engineering at scale since 2024.

Availability: Open to conversations about opportunities where I can make a meaningful contribution.
Email: matidesalegnece@gmail.com (personal), mathiwossilvio@gmail.com (work)
Book a thirty-minute call: https://cal.com/matiwos-desalegn-b2fps4/matidata, or through Calendly: https://calendly.com/matidata/30min

Full text of the public case studies, roles and services: https://matidata.ethioneural.com/llms-full.txt

## Pages

- [Home](https://matidata.ethioneural.com/): I help teams design, build and govern data platforms they can trust.
- [About Matiwos Desalegn](https://matidata.ethioneural.com/about): background, education, skills by evidence level, credentials and recommendations
- [Services](https://matidata.ethioneural.com/services): data architecture, engineering, governance and analytics engineering, with AI as a second discipline
- [Case studies](https://matidata.ethioneural.com/work): data platforms, AI engineering, open source and research, with the decisions behind each one
- [Data architecture: choices and principles](https://matidata.ethioneural.com/architecture): how I choose between warehouse, lake and lakehouse, ETL, ELT and CDC, and central or domain ownership, then the data-intensive systems principles, with examples from public code
- [Evidence, question by question](https://matidata.ethioneural.com/evidence): the questions hiring managers and clients ask, each answered with public work to check
- [Free checklists and guides](https://matidata.ethioneural.com/resources): a CDC pipeline checklist and a step-by-step CDC build guide, a platform choice scorecard and an AI readiness checklist, free to use
- [Talks and sessions](https://matidata.ethioneural.com/talks): talks Mati can give on change data capture, platform choices and AI readiness, and the sessions he has led
- [How I use AI](https://matidata.ethioneural.com/ai): where AI helps Mati's work, where he does not use it, and how its output is checked
- [Fintech](https://matidata.ethioneural.com/fintech): data platforms for digital lending, and the questions fintech data teams bring
- [Writing and research](https://matidata.ethioneural.com/writing): articles, and an Amharic language model paper in review
- [CV](https://matidata.ethioneural.com/cv): the CV, with data engineering and AI and ML versions
- [Now](https://matidata.ethioneural.com/now): what Mati is working on at the moment
- [Contact](https://matidata.ethioneural.com/contact): how to reach Mati

## Case studies

### Data architecture and platforms

- [World Bank CDC analytics: PostgreSQL to ClickHouse, orchestrated by Airflow](https://matidata.ethioneural.com/work/wb-cdc-analytics): A three-day take-home assessment, built on the World Bank's public Indicators API: PostgreSQL, Debezium into Redpanda, ClickHouse with dbt marts, Airflow, and Prometheus and Grafana, started with one command.
- [Kiva loan CDC analytics: PostgreSQL to ClickHouse, orchestrated by Dagster](https://matidata.ethioneural.com/work/kiva-microfinance-loan-cdc-analytics): A take-home technical assessment on Kiva's public loan API, extended since: PostgreSQL, Debezium into Redpanda, ClickHouse with dbt models and tests, Dagster, and Prometheus and Grafana with a reconciliation monitor.
- [CDC Starter: PostgreSQL to ClickHouse CDC in one command](https://matidata.ethioneural.com/work/cdc-starter): A template of the CDC pattern I use, small enough to read in one sitting: PostgreSQL, Debezium into Redpanda, ClickHouse and dbt in one command, with deletes, ordering, lag and reconciliation handled.

### AI engineering

- [Local RAG assistant with SSO, PII redaction and an audit log](https://matidata.ethioneural.com/work/local-rag-assistant): A document assistant that runs entirely on local hardware: nginx with TLS, a FastAPI gateway with JWT and Keycloak SSO, PII redaction before inference, AnythingLLM retrieval and Ollama, in Amharic and English.
- [Amharic OCR for Ethiopic script with HHD-Ethiopic and MMOCR](https://matidata.ethioneural.com/work/amharic-ocr-ethiopic-script): A full-stack OCR app for Ethiopic script: hierarchical detection, HHD-Ethiopic CTC recognition with confidence scores, multi-page PDFs and a Next.js canvas, plus an MMOCR research line with SATRN and DBNet++.
- [Natural language to SQL analytics agent over ClickHouse](https://matidata.ethioneural.com/work/nl-to-sql-analytics-agent): A prototype that turns plain-English questions into ClickHouse SQL with a Groq-hosted model and returns tables, charts and a short explanation in a React chat interface; built as a technical assessment prototype.
- [Hair virtual try-on: a React and FastAPI prototype](https://matidata.ethioneural.com/work/hair-virtual-try-on): A full-stack prototype: a React and TypeScript app to upload a photo and a hairstyle reference, compare before and after, and download the result, over a FastAPI endpoint whose model step is still a placeholder.

### Research

- [How small, how quantized? Tokenizer, size and quantization for Amharic SLMs](https://matidata.ethioneural.com/work/amharic-slm-quantization): An independent study of how tokenizer choice, model size and quantization each affect Amharic small language models on a single consumer GPU, with QLoRA recovery tested; submitted to a 2026 workshop, in review.

### Earlier work

- [Frontend of an HR and payroll ERP at HST Consulting](https://matidata.ethioneural.com/work/hst-payroll-frontend): My first industry role, where I built and maintained the frontend of a software-as-a-service HR and payroll system and integrated it with the backend team's REST APIs.
- [Student dropout prediction with a Streamlit dashboard](https://matidata.ethioneural.com/work/student-dropout-prediction): A two-week data science exercise that cleaned and explored a student records dataset, tested hypotheses about who drops out, and shipped the findings as an interactive Streamlit app.
- [Analytics and machine learning project set from 10 Academy](https://matidata.ethioneural.com/work/ten-academy-analytics-projects): A set of weekly end-to-end projects from the 10 Academy training programme, covering forecasting, fraud detection, credit scoring, a Telegram-sourced data warehouse and several exploratory analyses.
- [Cotton disease prediction with deep learning (final-year project)](https://matidata.ethioneural.com/work/cotton-disease-prediction-deep-learning): My final-year project classified cotton leaf diseases from photos with a convolutional network trained by transfer learning, served through a Flask web app and an Android app for use in the field.
- [Library database system in MySQL (database course project)](https://matidata.ethioneural.com/work/library-database-system): A course project that modelled a university library in MySQL, from the extended entity-relationship design through to the relational schema and the SQL scripts behind it.
- [WSU Guide, a campus navigation app for Wolaita Sodo University](https://matidata.ethioneural.com/work/wsu-guide-android-app): A native Android app that helps new students, visitors and staff find offices, dormitories, departments and services across the two Wolaita Sodo University campuses.

## Profiles

- [GitHub](https://github.com/matidesalegn)
- [LinkedIn](https://www.linkedin.com/in/matiwos-desalegn/)
- [Medium](https://medium.com/@mathiwossilvio)
- [X](https://twitter.com/Matdesalegn)
- [YouTube](https://www.youtube.com/@matiwosdesalegn)
- [ORCID](https://orcid.org/0009-0000-4912-2728)