Who I am

Yaroslav Vlasov — Software engineer. Saint Petersburg State University · Faculty of Mathematics and Computer Science · Modern Software Engineering, year 2. Saint Petersburg.

I build systems end to end — from the database schema and the deployment to the shaders. The site you are scrolling right now is the proof: every line of it is mine, from docker-compose to the last line of GLSL.

System pulse

• Nodes in the knowledge graph: 12
• Edges between them: 19
• System online: 4 hours
• Last deploy: August 18

Work

• Arcana

  A personal OS on my own server: planner, exam mode, knowledge graph, tree of goals.

  Next.js 16, PostgreSQL, Docker

• EDF scheduler

  Works out whether you will make your deadlines, and says plainly what to move.

  TypeScript, algorithms, tests

• Observatory

  A renderer for knowledge on bare R3F — the same one that drew this page.

  three.js, WebGL, shaders

More on GitHub

• C_Cpp — Practice in C and C++: algorithms and data structures.
• OOP_25-26 — лабы и проекты по 1 курсу ООП
• ConcoleCheck — Консольное приложение шашки
• Kotlin — Course projects in Kotlin.

Trajectory

1. July 2026 — Foundation on my own hardware

   Built: Next.js 16 + PostgreSQL + Docker Compose + Caddy with automatic HTTPS, layered data access (zod → repository → service).

   Learned: Defence in depth: a proxy redirect is UX, not a security boundary; the real check is the first line of every server action.

2. July 2026 — Backups that do not leak

   Built: pg_dump → gzip → gpg in a single pipeline, with no plaintext ever touching the disk; optional off-site upload of the encrypted dump.

   Learned: A backup without a tested restore is not a backup. And one intermediate file on disk undoes all of the encryption.

3. July 2026 — A planner with honest arithmetic

   Built: A priority engine and an EDF scheduler: it works out whether the deadlines are reachable and names what has to move. All of it under tests.

   Learned: An algorithm is worth exactly as much as it is trusted. So the scheduler does not just hand over a plan — it explains why a task ended up on top.

4. July 2026 — Estimates that learn

   Built: A timer and real minutes tracking; size estimates adjust to how the work actually goes.

   Learned: Feeding reality back into the model is the cheapest way to make a system more accurate without adding a single new form.

5. July 2026 — Exam mode

   Built: A 0–5 mastery model on top of the subject catalogue, separate calculation engines, import of a preparation plan.

   Learned: A mastery scale beats a “learned it” checkbox: it shows where you actually are, and where to go next.

6. July 2026 — Knowledge graph and the secret circuit

   Built: The knowledge graph as a projection layer over the data; the Tree of Goals in separate tables, with AES-256-GCM encryption of its contents.

   Learned: Secrecy holds better in architecture than in a filter: the private zone is simply never imported by the public one, and a guard test keeps it that way.

7. July 2026 — A public window

   Built: A whitelisted snapshot of the graph for the outside world, plus a separate showcase editor: draft, publish as a snapshot, roll back.

   Learned: Splitting “draft” from “published” costs one table and removes the one real fear — showing something by accident.

8. July — August 2026 — The “Skeleton of Knowledge” showcase

   Built: This scroll experience on bare React Three Fiber: a crystal skull, a knowledge graph inside the brain, a descent along the skeleton, custom dust and nebula shaders.

   Learned: In 3D almost every visible defect is not a wrong parameter but a wrong model. The jittery dust was not fixed by tuning — it was fixed by simulating velocities.

How this site works

• My own VPS, Docker Compose, Caddy with automatic HTTPS. Not a single external panel.
• Next.js 16 and PostgreSQL. The database port is not published at all.
• Eleven written architectural decisions — each one with its reasoning.

Full infrastructure diagram and architectural decisions

CV

CV on one page

Contact

• yarosvlasov@gmail.com
• GitHub
• Telegram