# Avinash Gosavi

> Staff Software Engineer at Cisco Meraki in the San Francisco Bay Area. Writes about memory for AI agents (privacy-preserving, brain-inspired retrieval), plain-language summaries of research papers, distributed systems and API design, Ruby on Rails, and local-first AI. Builds open-source tools such as VaultMem (zero-knowledge encrypted memory for AI agents) and Attractor (Hopfield-inspired memory engine).

Full text of every post in one file: https://www.avinashgosavi.com/llms-full.txt

## Start here

- [Paper TL;DR: Six Papers That Shaped How I Think About AI Memory](https://www.avinashgosavi.com/post/paper-tldr-six-papers-on-memory/): overview of the core papers behind agent memory, each in plain language
- [An Agent Memory Glossary](https://www.avinashgosavi.com/post/agent-memory-glossary/): 20 agent-memory terms defined simply
- [Research & Interests](https://www.avinashgosavi.com/research/): topics, papershelf with TL;DRs, newsletters
- [Projects](https://www.avinashgosavi.com/projects/): open-source projects with code and demos
- [About](https://www.avinashgosavi.com/about/): who Avinash is

## Series

### Building VaultMem

- [VaultMem — Building Encrypted Memory for AI Agents](https://www.avinashgosavi.com/post/vaultmem-encrypted-memory-for-ai-agents/): Why I built VaultMem, an encrypted memory library for AI agents where the platform running the agent cryptographically cannot read what users store.
- [VaultMem Part 2 — Multi-Modal Memory, Temporal Search, and Pluggable Backends](https://www.avinashgosavi.com/post/vaultmem-part-2-multimodal-temporal-backends/): VaultMem v0.2: multi-modal memory, temporal search, pluggable storage backends, and a first evaluation against the LoCoMo benchmark.
- [VaultMem Part 3 — Hiding PII from Your LLM Provider](https://www.avinashgosavi.com/post/vaultmem-part-3-pii-sanitizer/): VaultMem v0.3: a PII sanitizer that keeps personal details away from the LLM provider — the gap that encryption at rest doesn't close.

### Paper TL;DR

- [Paper TL;DR: Six Papers That Shaped How I Think About AI Memory](https://www.avinashgosavi.com/post/paper-tldr-six-papers-on-memory/): Short, plain-language summaries of six papers behind my AI memory side projects — from Hopfield's 1982 associative memory to user-owned memory for agents.
- [Paper TL;DR: Hopfield Networks Is All You Need — Attention Is a Memory Lookup](https://www.avinashgosavi.com/post/paper-tldr-hopfield-networks-is-all-you-need/): A plain-language walkthrough of Ramsauer et al. (2020): how modern Hopfield networks store exponentially many memories and why their update rule is transform...
- [Paper TL;DR: Complementary Learning Systems — Why Agents Need Two Kinds of Memory](https://www.avinashgosavi.com/post/paper-tldr-complementary-learning-systems/): A plain-language TL;DR of McClelland, McNaughton & O'Reilly (1995): the brain's fast and slow learners, catastrophic interference, and what it suggests for d...
- [Paper TL;DR: LoCoMo — How Good Is Your Agent's Memory, Really?](https://www.avinashgosavi.com/post/paper-tldr-locomo-long-term-memory-benchmark/): A plain-language TL;DR of the LoCoMo benchmark (Maharana et al., 2024) for very long-term conversational memory — what it tests, what it found, and how I use...
- [Paper TL;DR: Memory as Asset — Whose Memory Is It Anyway?](https://www.avinashgosavi.com/post/paper-tldr-memory-as-asset/): A plain-language TL;DR of Memory as Asset (Pan, Huang & Yang, 2026): moving AI memory from agent-centric to human-centric, its three principles and three lay...
- [Paper TL;DR: Sentence-BERT — The Quiet Workhorse Behind Semantic Search](https://www.avinashgosavi.com/post/paper-tldr-sentence-bert/): A plain-language TL;DR of Sentence-BERT (Reimers & Gurevych, 2019): why plain BERT is too slow for search, how siamese training fixes it, and why it still po...
- [Paper TL;DR: Hopfield (1982) — Memory as a Valley You Roll Into](https://www.avinashgosavi.com/post/paper-tldr-hopfield-1982/): A plain-language TL;DR of John Hopfield's 1982 paper on associative memory: energy landscapes, Hebbian storage, recall from partial cues and its limits — wit...

## All posts

- [An Agent Memory Glossary: 20 Terms in Plain Language](https://www.avinashgosavi.com/post/agent-memory-glossary/) (2026-10-19): Plain-language definitions of the terms that come up when building memory for AI agents — episodic vs. semantic memory, embeddings, HNSW, re-ranking, consoli...
- [Paper TL;DR: Hopfield (1982) — Memory as a Valley You Roll Into](https://www.avinashgosavi.com/post/paper-tldr-hopfield-1982/) (2026-10-18): A plain-language TL;DR of John Hopfield's 1982 paper on associative memory: energy landscapes, Hebbian storage, recall from partial cues and its limits — wit...
- [Paper TL;DR: Sentence-BERT — The Quiet Workhorse Behind Semantic Search](https://www.avinashgosavi.com/post/paper-tldr-sentence-bert/) (2026-10-18): A plain-language TL;DR of Sentence-BERT (Reimers & Gurevych, 2019): why plain BERT is too slow for search, how siamese training fixes it, and why it still po...
- [Paper TL;DR: Memory as Asset — Whose Memory Is It Anyway?](https://www.avinashgosavi.com/post/paper-tldr-memory-as-asset/) (2026-10-18): A plain-language TL;DR of Memory as Asset (Pan, Huang & Yang, 2026): moving AI memory from agent-centric to human-centric, its three principles and three lay...
- [Paper TL;DR: LoCoMo — How Good Is Your Agent's Memory, Really?](https://www.avinashgosavi.com/post/paper-tldr-locomo-long-term-memory-benchmark/) (2026-10-18): A plain-language TL;DR of the LoCoMo benchmark (Maharana et al., 2024) for very long-term conversational memory — what it tests, what it found, and how I use...
- [Paper TL;DR: Complementary Learning Systems — Why Agents Need Two Kinds of Memory](https://www.avinashgosavi.com/post/paper-tldr-complementary-learning-systems/) (2026-10-18): A plain-language TL;DR of McClelland, McNaughton & O'Reilly (1995): the brain's fast and slow learners, catastrophic interference, and what it suggests for d...
- [Paper TL;DR: Hopfield Networks Is All You Need — Attention Is a Memory Lookup](https://www.avinashgosavi.com/post/paper-tldr-hopfield-networks-is-all-you-need/) (2026-10-18): A plain-language walkthrough of Ramsauer et al. (2020): how modern Hopfield networks store exponentially many memories and why their update rule is transform...
- [Paper TL;DR: Six Papers That Shaped How I Think About AI Memory](https://www.avinashgosavi.com/post/paper-tldr-six-papers-on-memory/) (2026-10-18): Short, plain-language summaries of six papers behind my AI memory side projects — from Hopfield's 1982 associative memory to user-owned memory for agents.
- [Doing the Right Thing, Fast](https://www.avinashgosavi.com/post/doing-the-right-thing-fast/) (2026-04-25): Cleaning up after deprecating the first Catalyst Cloud Monitoring: splitting it into focused work streams and shipping it all in two weeks with AI help.
- [Live Audio Transcription on a $0 Server Using Whisper](https://www.avinashgosavi.com/post/live-audio-transcription-whisper/) (2026-04-12): Near real-time audio transcription on my own server with faster-whisper — no third-party APIs — and the open-source service I extracted from daylog.
- [What Should You Actually Use to Build Your Next Hobby Project?](https://www.avinashgosavi.com/post/hobby-project-stack/) (2026-04-12): What to use for your next hobby project in 2025: the zero-cost stack I chose for daylog — Cloudflare Pages, Fly.io, Neon, Upstash — and why.
- [How daylog Keeps Your Diary Private (Even from Itself)](https://www.avinashgosavi.com/post/daylog-security-architecture/) (2026-04-12): How daylog, a voice diary, keeps entries and its semantic search index encrypted, with nothing leaving the device to answer a query.
- [Attractor: Building a Memory Engine Inspired by How the Brain Works](https://www.avinashgosavi.com/post/attractor-memory-engine/) (2026-04-05): Attractor, a memory engine inspired by how brains remember — Hopfield-style retrieval by partial cues and context instead of keys or nearest neighbors.
- [VaultMem Part 3 — Hiding PII from Your LLM Provider](https://www.avinashgosavi.com/post/vaultmem-part-3-pii-sanitizer/) (2026-04-01): VaultMem v0.3: a PII sanitizer that keeps personal details away from the LLM provider — the gap that encryption at rest doesn't close.
- [Your AI Agent Shouldn't Be Able to Read Your Diary](https://www.avinashgosavi.com/post/vaultmem-your-ai-agent-shouldnt-read-your-diary/) (2026-03-28): A plain-language case for AI memory the platform can't read: if your assistant remembers your life, who else gets to see it?
- [VaultMem Part 2 — Multi-Modal Memory, Temporal Search, and Pluggable Backends](https://www.avinashgosavi.com/post/vaultmem-part-2-multimodal-temporal-backends/) (2026-03-28): VaultMem v0.2: multi-modal memory, temporal search, pluggable storage backends, and a first evaluation against the LoCoMo benchmark.
- [VaultMem — Building Encrypted Memory for AI Agents](https://www.avinashgosavi.com/post/vaultmem-encrypted-memory-for-ai-agents/) (2026-03-21): Why I built VaultMem, an encrypted memory library for AI agents where the platform running the agent cryptographically cannot read what users store.
- [Building Rena — An AI Health Companion with Gemini Live, Veo 2, and Google ADK](https://www.avinashgosavi.com/post/building-rena-ai-health-companion-gemini-live/) (2026-03-16): How I built Rena for the Gemini Live Agent Challenge 2026 — a voice-first health companion using Gemini Live, Veo 2 and Google ADK.
- [AI History Blog Generator](https://www.avinashgosavi.com/post/ai-history-blog-generator/) (2026-03-14): Building an agent that turns any historical topic into a Jekyll blog, milestone by milestone, using local models through Ollama.
- [Are These AI-Generated Tests Good?](https://www.avinashgosavi.com/post/reviewing-ai-generated-tests-istqb-framework/) (2026-02-28): A simple, repeatable way to review AI-generated tests, using ISTQB test-design ideas to spot the cases fast output tends to miss.
- [Building Your Website with Jekyll and GitHub Pages](https://www.avinashgosavi.com/post/building-your-website-with-jekyll-and-github-pages/) (2026-02-21): A practical guide to building a personal site with Jekyll and GitHub Pages using an AI-assisted editor: what to ask for, what to configure, where to read more.
- [Codex Skills: Baking Your Workflow Into the Tool](https://www.avinashgosavi.com/post/codex-skills-baking-your-workflow-into-the-tool/) (2026-02-17): How I stopped repeating the same instructions to my coding agent by encoding my workflow as Codex Skills — and when to use a skill vs. the agent.
- [I Fell in Love with Coding Again](https://www.avinashgosavi.com/post/i-fell-in-love-with-coding-again/) (2026-02-13): Why working with LLMs made coding fun for me again — talking through ideas, validating them fast, and two examples that made it click.
- [My SF Ruby 2025 Experience — Talks, Side Quests, Community & Inspiration](https://www.avinashgosavi.com/post/my-sf-ruby-2025-experience-talks-side-quests-community-inspiration/) (2025-11-27): My notes from SF Ruby 2025 — the talks I loved on Rails architecture, Ruby performance and AI, plus the side quests and community moments.

## Pages

- [Experience](https://www.avinashgosavi.com/experience/): work history
- [Talks worth watching](https://www.avinashgosavi.com/talks/): recommended conference talks with one-line reasons
- [Conferences & community](https://www.avinashgosavi.com/events/): events attended, hacked at or judged
- [Uses](https://www.avinashgosavi.com/uses/): hardware, home inference server, tools and hosting stack
- [Now](https://www.avinashgosavi.com/now/): current focus
- [Changelog](https://www.avinashgosavi.com/changelog/): what changed on the site recently
- [RSS feed](https://www.avinashgosavi.com/feed.xml)