Insights on Backend Engineering, AI/ML, and Full Stack Development
A deep dive into building a silver-standard dataset and using Direct Preference Optimization (DPO) to eliminate 'AI-isms' and master high-engagement writing.
Read "Engineering Virality: Fine-Tuning LLMs for Style Transfer using QLoRA and DPO"For years, social media algorithms have been like silent observers. They watched how long you lingered on a video, whether you hit the like button, or if you sent a reel to a friend. While these passive signals helped build the feeds we see today, they had a major flaw: watching something doesn't always mean you enjoyed it. Sometimes, we watch things out of habit, shock, or because we simply forgot to scroll. The Shift to True Interest: Meta engineers realized that to build a better Facebook Reels experience, they needed to stop guessing and start listening. This led to the creation of the User True Interest Survey (UTIS) model. Instead of just tracking your data, Meta began integrating direct feedback—asking users through mini-surveys if a video actually matched their personal interests.
Read "Why Facebook is Finally Asking What You Actually Want to See"A comprehensive guide to designing and implementing production-ready backend systems using Python, FastAPI, and modern cloud-native practices.
Read "Building Scalable Backend Systems with Python and FastAPI"A deep dive into RAG architecture, one of the most powerful techniques for enhancing LLM responses with external knowledge.
Read "Understanding Retrieval-Augmented Generation (RAG)"