Muhammad Fikri Wahidin

Developed and deployed the AI core of Mona, Netmonk's network-monitoring chatbot running in Telkom Indonesia's production environment. Engineered the NLU (Natural Language Understanding) intent-classification pipeline through three model generations (TF-IDF + SVM, fastembed + ONNX embeddings, and a fine-tuned IBM Granite multilingual embedding model), using LLM-based data augmentation (Gemini Flash, Gemma) to curate ~2,000 high-quality training samples from 5,000 raw. Built an end-to-end NER (Named Entity Recognition) pipeline in two generations, from rule-based matching to a custom SpaCy model trained on domain data, integrated across all four service layers. Engineered a TTS (Text-to-Speech) voice engine by fine-tuning a Piper ONNX model and designing a pre-generation + MinIO caching architecture that eliminated real-time inference latency. Shipped chatbot flows to Telegram and web-widget channels, managing the full dev, staging, and production deployment cycle. The custom spaCy NER model was deployed to staging.

Designed and implemented an ELK Stack-based security monitoring system within the Information Technology division to detect suspicious multi-device login activities across Telkom's infrastructure. Leveraged Elasticsearch for efficient log indexing and Kibana for comprehensive visualization, developing custom anomaly detection rules that identified high-risk IP addresses for proactive threat mitigation. Integrated Telegram API to deliver real-time alerts for critical security events, including simultaneous logins from 4+ devices within a one-minute window. Created interactive Kibana dashboards highlighting the top 10 users with abnormal login patterns, enabling faster decision-making for cybersecurity operations.

Completed 500+ hours of intensive technical training in machine learning fundamentals, TensorFlow, and end-to-end model deployment. Earned 35 professional certifications from Coursera and Dicoding, developing expertise in data preprocessing, model optimization, and production-grade ML workflows. Led development of "Harvest Scan" capstone project, implementing a CNN model using transfer learning with MobileNetV2 that achieved 95% validation accuracy for plant disease detection. Collaborated with Android and Cloud teams to integrate the model into a scalable application, employing Agile methodologies and tools including Git and Jupyter Notebook. Successfully delivered an ML solution addressing real-world agricultural challenges while mastering Python, TensorFlow, and cross-functional teamwork.

Completed a game development internship at Rumah Koding Indonesia, a coding bootcamp company based in Banjarbaru, during my second year of high school. Developed a 3D platformer game using Unity and C#, implementing game mechanics, user interfaces, and interactive elements. I also created a shooting game using Scratch, demonstrating versatility in development platforms while strengthening foundational programming concepts and game design principles.