Avi Chawla

Daily Dose of Data Science presents its material as continuing education for working practitioners. Its archive covers the path from data analysis and machine learning to retrieval, language-model serving, agent memory and production workflows. Individual explanations are paired with visual material or code so readers can connect a term with an operation. Chawla’s professional profile also records study at IIT (BHU), Varanasi from 2016 to 2020. His public writing is the central professional output documented by the publication.[2][5]

Explaining repeated computation

Chawla’s prompt-caching guide distinguishes the stable opening of a request from the conversation that changes on each turn. A provider can reuse computation for a matching prefix rather than processing those same tokens again. The article connects that mechanism to prompt organization: keep stable instructions and tool definitions together, then append changing messages and results. This gives a developer a concrete reason to examine context layout when an agent’s repeated calls become expensive.[3]

Choosing what a good answer means

His evaluation guide starts with a simple failure of text-overlap metrics: an answer can be correct while using different words from a reference. It compares methods that examine wording, semantic similarity, model judgments and an agent’s sequence of actions. The point is to choose a measure that reflects the task rather than treating one score as a universal definition of quality. That distinction matters when an application must both answer correctly and use tools appropriately.[4]