Tejas builds and operates a one-person AI factory: a local fleet of agents on his own GPU, a 28,000-file machine brain, and a self-hosted control plane that dispatches real work to real models. One laptop. Zero cloud bills.
This isn't a portfolio of AI-assisted homework. It's infrastructure: a working plant where agents get registered, dispatched, reviewed, and paid in tokens.





Self-hosted cockpit that dispatches real tasks to a registered agent fleet — queues, reviews, token accounting, autostart on boot.
control plane · productionA 28,000-file Obsidian vault where every AI on the machine reads identity, rules, reports and memory from one shared root.
the brain · 28,956 filesOllama, LocalAI and a LiteLLM tower put 300+ model endpoints behind one local door, with watchdogs keeping VRAM honest and API spend at zero.
local models · zero spendA living AI town of 19 workers with jobs, a GBA-style walking map, an auto-repair loop, and its own LoRA training pipeline.
the town · 19 workersEvery agent is registered, dispatched via a task queue, and audited after the job. These are the ones on active shift.
Shipped things, not intentions. Every line here exists on disk or on the internet.
The whole thesis, open-sourced: one operator, agent rings for surface, orchestration and brain, and the discipline to run it solo.
Rust daemon with a sidecar, word-by-word streaming, persona brains, QA'd live end to end. Built through three real bug hunts with a workflow that caught what review missed.
An 82-command MCP server resurrected from a dead plugin registry, then the fix written up and published so nobody else loses a week to it.
An automated episode factory: briefs, screenplays, renders, and a CEO gate script that decides what ships. Episodes 4 through 8 rendered and approved.
School-mode AI workstation with a think tool, now at v2.0. The daily driver that started the whole factory idea.
Right here, no download needed, no button — the project evidence is the Shift Log above. Want it on paper? There's a plain printable copy.
2nd Year, 3rd Semester — Indore, Madhya Pradesh. Building the business and law base to run a one-person company, not just work for one.
Physical stock audits and goods verification: counted inventory, reconciled counts against records, filled structured audit sheets, handed verified results to the client's CA for sign-off. Accuracy, honesty, on-site discipline.
August 2025 · June 2025. Employability and structured-thinking training on top of the degree.
Native-level Hindi–English translation, stock audit and inventory verification, basic Android flashing and rescue tooling, documentation and report writing. Comfortable with AI tools — this floor runs on them.