Terhemba Alumo

Terhemba AlumoAug 13, 2026

PowerPoint is more than creating attractive presentations—it can also play an important role in AI training and multimodal evaluation. As an AI Trainer, I apply presentation and visual communication skills to evaluate AI-generated PowerPoint slides for: • Content accuracy and relevance • Slide structure and logical flow • Layout, alignment, and readability • Visual consistency and presentation quality • Instruction-following and prompt adherence • Charts, images, tables, and other visual elements • Overall effectiveness and user experience This type of work helps AI systems better understand how to create clear, accurate, visually consistent, and professional presentations. Combining Microsoft PowerPoint, AI evaluation, data annotation, prompt interpretation, and quality assurance allows me to contribute to the development of AI systems that can produce more useful and professional content. I’m continuously developing my skills in AI Training, Multimodal Evaluation, Data Annotation, RLHF, Preference Data, and AI Quality Assessment. #AITraining #AITrainer #PowerPoint #DataAnnotation #AIEvaluation #MultimodalAI #RLHF #LLM #ArtificialIntelligence #MicrosoftPowerPoint

Terhemba AlumoOct 3, 2026

Today I completed the "Introduction to Agents" course on Mercor. The certificate recognized rigor, calibration, and clarity in AI agent evaluation, and those three ideas shaped everything I built. For the course, I created a customer response evaluation agent. Here's what it can do: ✅ Read the customer's original message and the support reply ✅ Score the reply on tone, accuracy, empathy, and clarity ✅ Check whether the customer's issue was actually resolved ✅ Flag weak spots, like robotic wording or missing steps ✅ Suggest a stronger version of the reply 🛠️ Built with: Claude Opus 5.5 What I learned along the way: 🔹 An agent is more than a prompt. It needs a clear role, a goal, and a way to reason through steps. 🔹 Your evaluation criteria make or break it. A vague rubric gives vague feedback, and a sharp one gives consistent results. 🔹 Testing is where the real learning happens. Every weak output showed me what to fix in my instructions. With my background in IT support and AI training and evaluation, this felt like a natural next step: moving from judging responses myself to building an agent that can do it at scale. This is just my first build, and I'm excited to build more. 🚀 What would you use an AI agent for? Tell me below 👇 #AIAgents #Mercor #ArtificialIntelligence #AITraining #CustomerSupport #LearningInPublic

Terhemba AlumoOct 19, 2026

Most customer service problems aren't hard. They're repetitive. "Where's my refund?" "Can I cancel my order?" "I'm locked out of my account." "What's the status of my payment? " Every one of these has a clear process, yet customers still wait in queues for answers. This is where AI agents are changing customer service. I recently built a customer support AI agent with Claude Code. It chats with customers and handles real requests: order cancellations, refunds, account restoration and payment ID tracking. When it can't resolve something, it escalates to a human instead of guessing. What I learned from building it, and from my time as a customer service agent at Vodafone and evaluating AI responses: ✅ Speed matters. Customers get answers in seconds, not hours. ✅ Tone matters. A correct answer delivered coldly still feels like bad service. ✅ Accuracy matters most. An agent that invents a refund policy is worse than no agent at all. ✅ Knowing its limits matters. The best agents know when to hand off to a person. AI won't replace good customer service. It will free human agents to focus on the complex, emotional cases where empathy really counts. #AI #CustomerService #AIAgents #LLM #ClaudeCode #CustomerExperience