codemigrator
Having the fact that banks run on ageing programming languages like COBOL which are a major bottleneck when it comes to upgrading or transitioning to new systems due to demanding nature of code bases as experienced engineers are required to read, analyse, generate and review code, not forget testing it, i built codemigrator for this specific issue. Codemigrator is a 5-agent AI pipeline that helps solve this issue in just about hours depending on the sisze of the codebase.This is what it does; 1. **Reads** legacy COBOL source files from the filesystem 2. **Understands** the business logic using extended AI reasoning 3. **Generates** production-ready TypeScript + M-Pesa Daraja 2.0 integration code 4. **Reviews** the output for OWASP Top 10 vulnerabilities and Daraja security requirements 5. **Tests** the output with a generated Jest suite that maps each test back to its COBOL paragraph
Elvis MuneneCloud 3ng1n33r
Cloud Guardian is an AI cloud engineer built with Google Gemini and ADK. It explores a Google Cloud environment, investigates resources, and looks for potential cost, security, and reliability problems. What makes it different is that it doesn't just report problems. It investigates the evidence, connects related resources, explains why something is a problem, prioritizes what should be fixed, and creates a remediation plan. With approval, it can also carry out and verify safe changes. Why Gemini? We chose Gemini because Cloud Guardian needs an AI that can reason, use tools, and understand different types of information. Gemini allows the agent to interact with cloud APIs, investigate problems across multiple steps, and even analyze things like architecture diagrams. In simple terms: Gemini is the brain, ADK is the agent framework, and Google Cloud provides the environment it manages.
James Kipsoi Study Agent on WhatsApp
A chat-operated study agent that gives students their university e-learning account inside WhatsApp. Students can ask what a topic is about, get real assignment questions, pull down this week's slides, get quizzed on them, and plan their day using natural language. Built using Google ADK + Gemini, deployed on Cloud Run, and accessible entirely through Twilio WhatsApp.
Samuel KangethePersonal Autonomous PC Agent (PAP)
PAP (Personal Autonomous PC Agent) is an AI assistant that lets users control and work with their computer using natural language or voice instead of manually navigating through applications, files, browser tabs, websites, and system settings. Users can ask PAP to perform tasks such as opening and managing applications, finding and working with files, controlling the system, searching the web, interacting with online AI services, and saving or restoring their working environment. For example, a developer can tell PAP to save their current development workspace and later ask it to resume the workspace, restoring the applications, project files, browser tabs, and terminal contexts they were working with. PAP doesn't simply execute fixed commands. Google Gemini understands the user's intent, plans the required steps, selects the appropriate tools and agents, executes the actions, verifies the results, and asks for clarification when the request is ambiguous. PAP can also learn useful user preferences over time, such as which browser profile or account to use for particular tasks. PAP moves AI from simply answering questions to actually getting things done on your computer.
Ian WabwireSemaBiashara
Most Kenyan shop owners — dukas, market traders, mama mbogas — can't tell you what their shop made this week. Not because they're bad at business, but because every bookkeeping tool asks them to learn software UX between serving customers all day. SemaBiashara flips that. A shop owner just talks — in Swahili, Sheng, English, or however they'd naturally describe their day — and Gemini transcribes, understands, and logs the transaction automatically, including slang money terms like "soo tano" (500) and "doo mbili" (2000). Ask "duka langu limefanya nini wiki hii?" and get back a real net-profit number, computed from sales minus cost of goods minus expenses, in one plain sentence. The agent is careful with money, not just fast. Ambiguous input gets a clarifying question instead of a guess. Mid-sentence self-corrections get caught. Credit sales to regular customers ("deni") are tracked as their own transaction type — a real, common pattern most bookkeeping tools ignore. Whatever the shop owner tells it, it records accurately and reliably, which is the actual hard problem here: getting language understanding right, not just wiring up a database. Built with Google's Agent Development Kit and the Gemini API, SemaBiashara runs across three interfaces — a text/voice ADK app, a Gemini Live-powered native voice app, and a custom branded chat UI — all sharing one ledger, so a transaction logged by voice on one surface shows up instantly when queried by text on another. Given the timeframe, we focused entirely on getting the core language understanding and logging right rather than spreading thin. Real M-Pesa reconciliation (Daraja API), actual SMS delivery, and USSD fallback for feature phones are natural next steps we scoped out but didn't build — the architecture is set up to add them without rework.
Timothy KipkoechORBIT — AI Project & Career Operations Agent
ORBIT is a multi-agent AI-powered Project and Career Operations platform designed to help individuals and teams plan, execute, monitor, and improve their projects while simultaneously supporting their professional growth. Powered by Google Gemini, ORBIT works as a team of specialized AI agents rather than a single chatbot. Each agent has a specific responsibility, while a central Project Commander coordinates them and determines which agent should handle a user's request. The Scrum Master Agent helps teams plan sprints, break down tasks, prioritize backlogs, conduct stand-ups and retrospectives, identify blockers, and monitor project health. The Team Coordinator Agent helps assign responsibilities, manage dependencies, coordinate team members, and identify tasks that are falling behind. ORBIT also acts as a Career Intelligence Assistant. Its Career Scout Agent searches for current Project Management internships, associate and junior roles, project coordinator positions, and other relevant opportunities that candidates in Kenya can apply for, including remote and hybrid opportunities. Users can upload their CV, which is analyzed by the CV Coach Agent. The agent identifies missing skills, weaknesses, ATS issues, missing achievements, and gaps between the user's current profile and the requirements of real Project Management positions. The Learning Coach Agent then converts those skill gaps into a personalized learning roadmap. It recommends free learning resources, practical projects, certifications, and portfolio activities while tracking the user's progress toward becoming a stronger Project Manager. The goal of ORBIT is to create a continuous loop: Project → Progress → Skills → Career Opportunities → Learning → Improvement Instead of simply telling users what to do, ORBIT helps them plan the work, coordinate people, track progress, discover opportunities, identify skill gaps, and continuously improve.
Alois GitauGeospatial Environmental Monitoring Agent(GEMA)
It's a Geospatial Environmental Monitoring Agent built using Python and Google's Agent Development Kit (ADK) framework. It does mining detection,inland water body mapping,maritime surveillance and spatial resolution.
Imbaya AmonPredict next period
This project helps users predict their menstrual cycle. User inputs their cycle length and last day of their previous cycle,the agent then predicts the upcoming one. This ensures user is fully prepared for their menses and assists in menstruation tracking
Janice Wangaremulti_tool_agent
A Gemini-powered agent built with Google's Agent Development Kit (ADK) that helps users — especially farmers — make weather-informed decisions. It answers questions about current weather, local time, and 3-day rainfall forecasts for any city worldwide by calling live, free APIs (Open-Meteo for weather/geocoding, timezonefinder for timezone resolution). Beyond text queries, it accepts an uploaded photo of a crop or field and combines Gemini's native vision capabilities with the live weather/rainfall data to give practical, specific farming advice — for example, identifying the growth stage of a maize crop in a photo and recommending planting timing based on the coming days' rain forecast. The agent autonomously decides which tool(s) to call based on the user's question, and reasons over both the tool outputs and the visual input to produce a single, coherent recommendation.
Samuel KaranjaZetech University Feedback Portal
The Zetech University Feedback Portal is an AI-powered platform designed to improve communication between students and university administrators by transforming student feedback into actionable insights. Traditional feedback systems often collect concerns but lack the ability to analyze large amounts of information, identify recurring issues, and support timely responses. This platform addresses that challenge by integrating Google's Gemini API with the Agent Development Kit (ADK) to create a multi-agent AI system. The solution includes specialized agents such as the Feedback Assistant Agent, which helps students submit meaningful feedback; the Issue Classifier Agent, which uses Gemini's language understanding capabilities to categorize and prioritize concerns; the Admin Insights Agent, which analyzes feedback patterns and generates institutional insights using connected tools; and the Admin Response Generator Agent, which assists administrators in creating effective responses to student issues. The platform was built using React for the frontend, Flask for backend services, MySQL for data storage, and Google Gemini with ADK for intelligent agent orchestration and tool-based workflows. By combining AI reasoning with real application data, the Zetech University Feedback Portal transforms feedback from isolated messages into meaningful intelligence that helps institutions listen better, understand faster, and respond more effectively. Demo Website: https://zetech-feedback-portal.vercel.app/ Test Credentials: Student Account: Email: [email protected] Password: zetech123 Admin Account: Email: [email protected] Password: zetech123
Eliezer BarackSales Coach AI
I built SalesCoach AI, a sales consultant powered by Google's Gemini API and the Agent Development Kit. It not only just answer questions. It decides whether to teach you a framework, craft a custom pitch, handle a specific objection, or become a tough customer so you can practice. For example, let's say a user (most probably a salesperson) tells the agent, 'My biggest challenge is convincing customers.' The agent doesn't just reply—it reasons across the conversation, remembers what you said, and might pull the SPIN framework, then roleplay as a skeptical buyer to test you, then analyze a real customer email you paste in.
Ryan WelimeLead Ranking AI Agent
The Problem: B2B sales teams waste hours manually vetting lead lists against ICPs, often relying on outdated database dumps.The Solution: An automated ADK pipeline that conducts live web research, scores fit, and guarantees full source-URL traceability for every score. Why Gemini & ADK: We leverage Gemini's search grounding capability combined with ADK's SequentialAgent workflow to strictly separate unconstrained web research from structured, validated output generation.
Morgan Muchira