Built the red-flag override system for the health triage assistant —
urgent symptoms like difficulty breathing now immediately trigger “High”
guidance instead of just adding up points. Hit and fixed a real bug
along the way (a variable used before it was declared). Also rewrote
this project’s README myself after getting feedback that my earlier
drafts sounded too AI-generated.
.png)
Built the red-flag override system for the health triage assistant —
urgent symptoms like difficulty breathing now immediately trigger “High”
guidance instead of just adding up points. Hit and fixed a real bug
along the way (a variable used before it was declared). Also rewrote
this project’s README myself after getting feedback that my earlier
drafts sounded too AI-generated.
.png)
Built the core risk-calculation logic for Flood Risk Index today. Ran
into a real problem first: the map tiles I’d originally set up (CartoDB
dark tiles) started requiring an API key I don’t have — the whole map
was covered in “API KEY REQUIRED” watermarks. Fixed it by switching to
OpenStreetMap’s standard tiles (permanently free, no key) and using a
CSS color-inversion trick to keep the dark navy look I wanted.
Then wrote calculateRisk() myself — it takes real elevation and rainfall
data for any point someone clicks on the map, and combines them into a
Low/Moderate/High risk score using simple, transparent rules (low
elevation + high rain chance = higher risk). Also wrote a full README
explaining the project, and being upfront that this uses a rule-based
formula rather than a trained ML model on satellite imagery — that’s a
deliberate scope decision (real, explainable results now) rather than an
unfinished feature, with the ML version listed as future work.
.png)
Built the core risk-calculation logic for Flood Risk Index today. Ran
into a real problem first: the map tiles I’d originally set up (CartoDB
dark tiles) started requiring an API key I don’t have — the whole map
was covered in “API KEY REQUIRED” watermarks. Fixed it by switching to
OpenStreetMap’s standard tiles (permanently free, no key) and using a
CSS color-inversion trick to keep the dark navy look I wanted.
Then wrote calculateRisk() myself — it takes real elevation and rainfall
data for any point someone clicks on the map, and combines them into a
Low/Moderate/High risk score using simple, transparent rules (low
elevation + high rain chance = higher risk). Also wrote a full README
explaining the project, and being upfront that this uses a rule-based
formula rather than a trained ML model on satellite imagery — that’s a
deliberate scope decision (real, explainable results now) rather than an
unfinished feature, with the ML version listed as future work.
Solved the AI backend problem in EthioFarm AI. The chat previously called
an external AI API directly from the browser, which isn’t realistic to
deploy without a backend server and paid API key. I rebuilt it as a
self-contained rule-based assistant instead: it detects what a farmer is
asking about (planting, irrigation, or rain) via keyword matching, then
combines that with real live weather data to generate a specific answer.
No external dependency — works identically whether local or hosted.
I wrote the core logic (detectIntent and generateAnswer functions) myself,
working through conditional statements step by step rather than just
pasting code. Tested it end-to-end — asked real questions and got correct
answers based on live weather data for a real location.
Also wrote a full README documenting the problem, solution, tech stack,
Solved the AI backend problem in EthioFarm AI. The chat previously called
an external AI API directly from the browser, which isn’t realistic to
deploy without a backend server and paid API key. I rebuilt it as a
self-contained rule-based assistant instead: it detects what a farmer is
asking about (planting, irrigation, or rain) via keyword matching, then
combines that with real live weather data to generate a specific answer.
No external dependency — works identically whether local or hosted.
I wrote the core logic (detectIntent and generateAnswer functions) myself,
working through conditional statements step by step rather than just
pasting code. Tested it end-to-end — asked real questions and got correct
answers based on live weather data for a real location.
Also wrote a full README documenting the problem, solution, tech stack,
Built the dynamic weather horizon (updateHorizon) to switch sky gradients and cloud visuals in real time.Implemented the async loadWeather function to fetch forecasts and handle status updates.Coded event listeners for the AI chat so users can send messages via button clicks, the Enter key, or suggestion chips.Cleaned up script typos and DOM element references to make everything run smoothly.
Built the dynamic weather horizon (updateHorizon) to switch sky gradients and cloud visuals in real time.Implemented the async loadWeather function to fetch forecasts and handle status updates.Coded event listeners for the AI chat so users can send messages via button clicks, the Enter key, or suggestion chips.Cleaned up script typos and DOM element references to make everything run smoothly.
Worked on the loadWeather(place) asynchronous function to manage fetching location data and updating current weather properties.Integrated fetchweather, updateHorizon, and recommendation building logic into the main loading sequence.Refined status message handling and location string formatting for the user interface.
Worked on the loadWeather(place) asynchronous function to manage fetching location data and updating current weather properties.Integrated fetchweather, updateHorizon, and recommendation building logic into the main loading sequence.Refined status message handling and location string formatting for the user interface.
Worked inside index.html on the updateHorizon(mood) JavaScript function.Refined conditional checks for weather states like clear, partly cloudy, and overcast skies.Fixed a spelling typo in the mood condition check to ensure the dynamic sky gradients and SVG cloud elements render smoothly.
Worked inside index.html on the updateHorizon(mood) JavaScript function.Refined conditional checks for weather states like clear, partly cloudy, and overcast skies.Fixed a spelling typo in the mood condition check to ensure the dynamic sky gradients and SVG cloud elements render smoothly.
Started EthioFarm AI — an AI climate assistant for Ethiopian farmers. So
far I’ve built the core structure: a weather-fetching system using real
location + forecast data, a visual “horizon” that changes based on actual
weather conditions, and the beginning of an AI chat feature. Still
debugging the weather display right now, and I’m working through.
Started EthioFarm AI — an AI climate assistant for Ethiopian farmers. So
far I’ve built the core structure: a weather-fetching system using real
location + forecast data, a visual “horizon” that changes based on actual
weather conditions, and the beginning of an AI chat feature. Still
debugging the weather display right now, and I’m working through.
Built the full SMS parsing pipeline for TrustPay — the Android app listens
for Telebirr and CBE Birr payment SMS, parses the amount/payer/sender, and
pushes it live to Firebase. This week I focused on securing it: added
Firebase Authentication so the database isn’t wide open, removed an
accidentally-committed config file with API keys, and rebuilt the web
dashboard to group transactions by day with separate running totals for
Telebirr vs CBE. Dashboard is now hosted live via GitHub Pages with a
password gate. Next: polish the README and start user-testing with the
actual shop.
Built the full SMS parsing pipeline for TrustPay — the Android app listens
for Telebirr and CBE Birr payment SMS, parses the amount/payer/sender, and
pushes it live to Firebase. This week I focused on securing it: added
Firebase Authentication so the database isn’t wide open, removed an
accidentally-committed config file with API keys, and rebuilt the web
dashboard to group transactions by day with separate running totals for
Telebirr vs CBE. Dashboard is now hosted live via GitHub Pages with a
password gate. Next: polish the README and start user-testing with the
actual shop.