"Yara" means friend in Hindi. I named it deliberately.
I read a lot. Articles, threads, research papers, investor memos, engineering posts. Probably 30-50 pieces a week across AI, enterprise software, content systems, and organizational design. For years, the pattern was the same: read something interesting, highlight a few lines, forget where I saw it three days later.
Yara changed that.
Yara is a personal knowledge system I built on top of Claude Code. It maintains a structured wiki of everything I read, tracks the people and ideas I follow, and helps me turn raw research into writing.
The architecture is simple. Three layers:
On a typical day, Yara might:
Beyond research and writing, Yara maintains my content strategy. It knows my four content pillars, tracks which ones are underserved, builds out a publishing calendar based on what's drafted and what's missing, and flags when I'm falling behind. It's the accountability partner that never forgets a deadline. When I tell it I want to publish three times a week across LinkedIn, Substack, and X, it holds me to that, surfacing gaps before they become missed weeks.
It also helps with side projects occasionally. Research for a real estate evaluation. A competitive brief. Anything where structured thinking benefits from having a knowledge base underneath it.
Yara handles the part I'm bad at: remembering where I read something, connecting it to three other things I read last week, keeping a structured record that compounds over time.
I handle the part it's bad at: deciding what matters, forming an argument, knowing when a piece is done.
Yara runs as a set of Claude Code skills, each one handling a different workflow: ingesting sources, maintaining the wiki, scanning for new reading material, drafting content, scoring drafts against quality guidelines, managing a publishing calendar.
The wiki itself is plain markdown. Every page is readable in a text editor, browsable in Obsidian, and published to yara.vishalsood.com as a static site. No database. No proprietary format. If the AI layer disappeared tomorrow, the knowledge base would still be useful.
That's a design choice, not an accident. The value isn't in the tooling. It's in the accumulated connections between sources.
The pattern Yara follows isn't original to me. Andrej Karpathy wrote about building an LLM wiki. Several people built their own versions independently, in the same week, without coordinating. The architecture is converging because the problem is universal: we all read more than we can remember, and search is a poor substitute for structured knowledge.
I built Yara because I needed it. I'm sharing it because the approach generalizes. Whether you're a product leader trying to synthesize market signals, a marketer tracking competitive moves, or an engineer keeping up with a field that moves weekly, the core question is the same: how do you turn reading into thinking at scale?
Yara is my answer. Yours will look different. The point is to build one.
I write about what I learn from building Yara on Substack and LinkedIn.