Work
Technical posts I wrote and produced for company and founder accounts, translating product ideas, research, and releases into platform-native stories.
RAG (Retrieval Augmented Generation) vs. Memory These two terms are often confused, but they play very different roles. RAG grounds an LLM with external knowledge. A query is embedded, matched against a vector store, and the top chunks are injected into context. Memory is broader. It stores, processes, and later retrieves distilled knowledge that supports decisions.
AI Agent Memory isn't just bigger context windows. It's how agents actually learn, adapt, and evolve. Context windows only let a model see more of the conversation at once. Memory is different: it is what the agent keeps, updates, and recalls to guide decisions.
Most AI agents today are still stateless. That’s why they don’t evolve over time. A stateful agent learns from interactions, stores what matters, and recalls it when needed.
How Mem0 works under the hood A technical walkthrough of the memory pipeline: context retrieval, extraction, memory updates, and storage.
Agent memory is still discussed as a retrieval problem. “Memory in the Age of AI Agents” argues that this framing is incomplete and treats memory as an evolving state inside the agent loop.
The hardest part about giving AI agents memory? Making them understand time. Most agents know what happened, but not when it happened.
New cookbook: Gemini 3 + Mem0 MCP A step-by-step guide to adding persistent, structured memory to a Gemini workflow using the Model Context Protocol.
Mem0 v1.0.0 is live. Faster memory retrieval, new integrations, and major performance upgrades across SDKs.
Two years ago, it was just my co-founder and me. I had just moved from India for the first time, and he had just left his job at Tesla. Today, we're building the future of AI memory with Mem0.
Last week, the entire Mem0 team went to Goa for our first offsite. Being a remote team based across India and the US, everything we knew about each other came from online interactions. Meeting in person genuinely felt different.
The next big leap in AI won't come from bigger models. It'll come from agents that remember. Six months ago, everyone building AI agents cared about getting them to take action through tool calls, planning, and execution. Memory was always an afterthought. Now that's changing. The best teams are starting with memory and designing around what needs to be remembered, not just what needs to be done.
Everyone is racing to build agents, but we're ignoring the most human part of intelligence. In his chat with Dwarkesh, Andrej Karpathy put it in one line: AI agents today don't have continual learning. You can't tell them something and expect them to remember it.
Bigger context windows can't eliminate the need for memory. A context window defines how much information a model can see at once. Expanding it to 200K or even a million tokens helps with short-term recall, but it doesn't help the model remember. Once the window slides, the model forgets everything outside it. It can't build on prior sessions, refine preferences, or evolve its understanding.
I liked how Sam Altman framed memory on Big Technology Podcast. What stood out was how early this still is. Most AI agents today are still largely stateless. They perform well within a session, but once it ends, context resets.
