In Ethiopian secondary schools, students consistently face severe academic barriers due to fragmented textbooks, lack of centralized curriculum materials, uneven educational resource access, and absence of structured self-testing tools across Grades 9-12.

Built a comprehensive high school educational platform paired with an automated Telegram bot delivery engine and high-performance microservices. Serves complete curriculum notes, digital textbooks, weekly credit-based mock exam sessions, and automated scoring analytics. Architected with custom PHP on a Laravel-based architecture, HTML & Tailwind CSS, a high-throughput REST API, an integrated custom-built AI Gateway, MySQL on cPanel, and Node.js & Fastify services to handle concurrent student submissions without expensive cloud scaling.

In Ethiopia and East Africa, Telegram is the dominant digital business and community channel, yet Telegram advertising remains plagued by manual 1:1 negotiations, unpredictable pricing, unverified engagement claims, and lack of secure local payment settlement.

Architected and launched TeleAds (teleads.tech), an end-to-end managed Telegram advertising marketplace. Advertisers purchase curated, category-specific channel packages with frozen order snapshots and escrow verification. Once creative review passes, the TeleAds bot automates top-placement post publishing and deletion monitoring, while a performance engine dynamically weights creator payouts in ETB based on verified impressions, views, and engagement rates.

Local community pharmacies routinely suffer financial losses and safety hazards from untracked medication expirations, manual inventory errors, and unverified dispensing in paper registers.

Engineered a centralized ERP-level pharmacy management system featuring automated FEFO (First-Expiry First-Out) stock depletion, real-time expiry warning thresholds, multi-branch POS terminal transactions, purchasing orders, and role-based access control.

Modern text and note-taking apps have evolved into sluggish, multi-gigabyte Electron bloatware requiring mandatory accounts, cloud sync, and network connectivity, while consuming massive RAM just to edit plain text files.

Built Markit with Go and Wails v2: a fast, distraction-free desktop Markdown editor for Windows & Linux. Delivers instant startup (<1s), side-by-side split-view live preview, dynamic document outline (H1-H6), fsnotify external file watching with smart conflict resolution (Reload, Save, Keep), native OS file dialogs, zoom controls (50% - 250%), quick search (Ctrl+F), and CLI startup support, with 0 Electron, no server, and no account.

Navigating the expansive Adama Science and Technology University (ASTU) campus is challenging for new students and visitors due to complex building codes, unmapped internal pedestrian paths, and lack of building floor directories.

Engineered an interactive campus navigation web application for ASTU using Next.js 16, TypeScript, Leaflet, and OSRM. Features high-resolution Esri satellite imagery, Dijkstra-powered foot-path routing with synchronized turn-by-turn guidance, building search with instant flyTo camera centering, floor-by-floor room directories, live GPS hardware tracking, and a built-in map digitizer canvas (/digitize).

Students and schools lack structured, real-time assessment feedback, standardized national mock examinations, and scalable digital preparation infrastructure.

Built a dedicated examination preparation platform delivering timed test simulations, comprehensive performance analytics, and structured subject drills. Engineered with custom PHP on a Laravel-based architecture, HTML, Tailwind CSS, high-speed REST API services, custom AI Gateway integration, and relational MySQL database storage on cPanel.

Developers integrating AI face single-point-of-failure risks, frequent provider rate limits (429), regional outages, and fragmented API schemas requiring custom adapters for each AI company.

Architected an open-source AI orchestration gateway providing one unified endpoint for 9 AI providers (Gemini, Groq, Mistral, OpenRouter, OpenCodeZen, Zydit, Zai, Agnes, Nvidia). Features automatic failover cascades, provider load balancing, API token authentication, and a 15 requests/minute per-IP rate limiter.

Standard Telegram automation relies on the Telegram Bot API or webhooks, which require separate bot handles, public server IPs, port forwarding, and cannot act directly on personal 1:1 user direct messages.

Engineered an intelligent state-of-the-art auto-reply agent for 1:1 private messages using personal Telegram account MTProto (GramJS). Dynamically unifies Telegram Cloud history with session streams (8-15 turns), simulates natural human typing delays, converts Markdown to native Telegram HTML, and enforces a strict gatekeeper firewall blocking sensitive financial and credential keywords (e.g. CBE, Telebirr, passwords).

Standard LLMs hallucinate when asked domain-specific questions about private documents, while naive keyword search fails to capture semantic meaning when queries and texts do not share identical vocabulary.

Built a production-grade Retrieval-Augmented Generation (RAG) system with a multi-stage pipeline: automated document parsing with metadata extraction, semantic text chunking (~2200 chars with 350-char overlap), and instruction-based high-dimensional vector search using the BGE-large-en-v1.5 embedding model.