Akshay Pachaar is an AI educator and cofounder of Daily Dose of Data Science. He publishes technical explanations and runnable examples about language models, retrieval and agent systems, drawing on earlier work at Lightning AI, TomTom and HERE Technologies.[3][1][6]
Start with Agent Memory Architecture Guide →
Education
Integrated master’s in mathematics and Bachelor of Engineering in Electrical and Electronics Engineering
[1]Links
X 
GitHub LinkedIn Google Scholar Links Cofounder · Daily Dose of Data Science
Career
Pachaar’s public biography records an integrated mathematics master’s and electrical-and-electronics engineering degree at BITS Pilani. He applied machine learning to maps and advanced driver-assistance systems at TomTom, after earlier work at HERE Technologies. Following his time at Lightning AI, he joined Avi Chawla as cofounder of Daily Dose of Data Science in October 2024. They develop newsletter issues, technical deep dives and educational material together.[1][3]
Contributions
Explaining memory as a system design choice
“Build Agents that never forget” introduces memory through progressively richer storage approaches, from a Python list and files to vector retrieval and combined graph-and-vector systems. It explains why repeatedly resending a conversation is different from maintaining persistent, selectively retrievable information. The useful contribution is the design comparison: readers can connect an agent’s need for past facts to storage and retrieval choices instead of treating a larger context window as the complete solution.[4]
Guides that connect configuration to runnable examples
Pachaar’s guide to the .claude folder distinguishes repository configuration from personal machine settings and walks through instructions, commands, skills, agents and permissions. His Lightning AI research tutorial moves from explanation to an application: a researcher, analyst and report-writer pipeline turns a question into a cited Markdown report, with optional document context and a Streamlit interface. These are educational implementations built with existing models and libraries, rather than claims of having invented the underlying model or orchestration framework.[5][6]
Tools, writing & talks
Writing, research & talks
Start here
A first-principles comparison of conversation history, files, vector retrieval and graph-based approaches to agent memory.
Start here if you can call a model API but are unsure what an agent should retain between requests. The guide builds the storage choices step by step; programming familiarity helps with implementation, not the initial comparison.
Author[4]Original sourceA thread on Kimi K3's delta attention, which replaces the growing KV cache with a single matrix, letting the open frontier-scale model hold a million tokens of context without the usual memory cost.
Catalog workOriginal sourceA guide to repository and personal configuration, instructions, commands, skills, agents and permissions.
Author[5]Original source
Tools
- Multi-Agent Deep Researcherlightning.ai
A runnable teaching application with research, analysis and report-writing agents, web retrieval and a Streamlit interface.
Tutorial author; uses existing models and orchestration libraries[6]