A dashboard for my live AI projects and knowledge systems.
Open the parts that matter fastest: live system status, active projects, learning tools, and the knowledge vault behind them.
This page is the front door. It shows what is running, what changed recently, and where to go next without making you learn the internal map first.
Connect+Act (reads knowledge, implements improvements)
Research + Synthesis (builds knowledge library)
Latest website-facing telemetry snapshot available on the public surface.
Open live status
Check instance health, machine status, active loops, and the latest system pulse in one place.
How the system works
See the operating model behind the site: prompts, agents, local models, and the workflow that turns ideas into shipped work.
Open the knowledge vault
Go to the vault dashboard for notes, retrieval, agent memory, and backend activity.
Resume active project work
Jump into production apps, showcases, and the surfaces that are actively being shipped or refined.
Enter learning mode
Switch from operating the system to using it: review notes, follow paths, and turn the vault into actual recall.
Browse topic clusters
Use domain-level entry points when you know the territory but not the exact note, post, or app.
Autonomous changes tracked across the system, with website-facing telemetry exposed instead of hidden in private tooling.
Concepts, links, and graph structure are part of the backend, not just loose notes sitting in folders.
The vault is not just storage. It is an active research corpus with prompts, synthesis, and retrieval structure baked in.
A Neural Router for My Knowledge Vault
I built a 15-agent routing system with a 7-layer scoring formula, a self-improving optimizer, and a hygiene scanner. The real insight wasn't any single layer — it was that sophisticated infrastructure without a discoverability layer rots on the shelf.
I Ran Autoresearch on My Own Website
Karpathy's autoresearch optimizes train.py against validation loss. What happens when you point the same pattern at a production website? 9 iterations, a 68% bundle reduction, 120 new internal links, and a real-time Convex backend — all discovered autonomously.
The Science of Spaced Repetition: Why Spacing Your Learning Matters More Than You Think
Discover why spaced repetition is the most empirically supported learning technique in cognitive psychology, how it works in your brain, and how the vachsark learning engine uses it to maximize your retention.
A live feed of the latest autonomous improvement lands here whenever a public-facing change ships.
Prompts and questions
Work begins as goals, questions, and tasks — the raw intent layer behind research, coding, and operations.
Agents and loops
Different agents route, research, implement, review, and monitor. Some loops generate knowledge; others turn it into changes.
Knowledge vault
The Obsidian vault stores notes, links, and synthesis. It works as a searchable memory system instead of a pile of documents.
Local models and infra
Local inference, GPU experiments, heartbeat jobs, dashboards, and scripts keep the backend moving and observable.
Public pages
The website, dashboards, learning pages, and project showcases are the visible layer people can actually browse and use.
Every file, function, import, and call in this site, parsed into a knowledge graph. Click a node to inspect callers and callees. Toggle Show functions to drill from architecture into implementation.
Indexed with gitnexus · regenerated on each build
Production apps and current builds
The shipped layer: software, tools, and experiments that already have a real interface or real users.
Research posts, implementation notes, and public thinking
The explanation layer: why systems were built, what changed, and what I learned while making them work.
Structured learning paths powered by the same knowledge vault
Review notes, follow learning paths, and turn a large vault into something you can actually remember and use.
Website as front door
The vault stays deep and powerful, but the site becomes the human interface: fewer decisions up front, clearer next steps, less cognitive drag.
Route by intent
When tired, the first question should not be which note. It should be whether you are checking status, resuming work, learning, or browsing.
Receipts over abstraction
Stats, running surfaces, recent writing, and visible architecture do more than broad claims. They prove the world is real.
The goal is not to browse everything. The goal is to know what this system is, what is alive right now, and where to go next.
At this scale, the system should route attention instead of demanding that you remember the internal shape of the vault. This version of the page makes that more explicit: clearer entry points for new visitors, a more useful operator snapshot for me, and stronger proof that there is a real knowledge-and-agent world underneath the site.