Nathaniel Abegunde Resume

• Fine-tuned multiple time-series LLM architectures (Llama 3.2 1B, ChatTS) on 4.4M satellite conjunction events using Unsloth, PEFT and QLoRA techniques with 8GB VRAM constraints.
• Optimised SpaceTube's post ranking algorithm, reducing response time from 257ms to 109ms (58% improvement) with TypeScript.
• Implemented real-time error monitoring system using Google Chat webhooks to alert on payment and API errors with FastAPI and NestJS.
• Developed RAG system using HuggingFace embeddings and ChromaDB retrieval to improve responses for space domain queries.

• Implemented ETL pipeline for livestock data processing, extracted 10+ hours of real-world audio and transformed through various denoising techniques for ML classification.
• Experimented with band-pass filtering and autoencoder-based denoising to optimise audio quality, achieving over 90% model convergence.

• Developed ML pipeline that scraped and analysed over 75,000 customer reviews for sentiment analysis.
• Using Random Forest and XGBoost, customer booking prediction achieved 85% accuracy.

NOTE: This was a virtual internship, and the duration for the internship is not in actual reality the time it took me to complete the tasks required to obtain a certificate.