I build production AI systems and own the architecture end to end.

Twenty-five years in software, most of it in private repositories. This account was created in July 2026 as a deliberate portfolio: each repo is either a sanitized cut of a system I actually run, or a study built end to end on public data, and every README says which.

That span includes running my own company. The private originals behind these repos run on real money and private data, and several have been in continuous use for months. Each repo below shows the design, runnable code, and clips of the real thing.

Every repo on this account carries one of two labels:

• Production extract  ·  A sanitized public cut of a private system in real use. The architecture and method are real; data, names and some components are stand-ins, and the README lists which is which.
• Portfolio build  ·  Built end to end for this portfolio on public data. Everything here runs as shown; there is no private original behind it.

Proving Ground  ·  Portfolio build
Generative and retrieval AI over vast game data. An LLM drafts warships from one sentence without ever seeing the validation rules; a grounded assistant reads 149,288,931 rows of synthetic combat telemetry in indexed Postgres and cites the exact rows or refuses. One typed data contract validates every record, on ingest and on generation.

PARALLAX  ·  Portfolio build
A backtest ends in one number and a user reads it as a fact about the past. Three measurements of the room between that number and the market it describes, on public exchange data anyone can re-derive.

Concierge  ·  Production extract
Drafts grounded replies to customer emails over typed MCP tools, into Gmail Drafts behind a policy gate and human review. Never auto-sends; prompt-injection attempts are quarantined.

Quant Watchtower  ·  Production extract
Read-only MCP operations console over a live 24/7 algorithmic trading fleet. Sanitization is enforced in the data layer, so the strategy never reaches the model.

Cortex  ·  Production extract
Self-hosted knowledge-AI platform: an Obsidian vault wired for AI agents through an MCP layer, Docker sync, and Python tooling. Ships a runnable clean-room linter.

Dev System  ·  Production extract
The method behind all of these: bounded efforts, standing invariants, a per-change audit trail, acceptance discipline. Validated on a live production system for two-plus months.

Celestia  ·  Production extract
Persona-driven assistant over a private Obsidian vault through a typed MCP filesystem core. Reads before it answers; every write routes to its owning note.

Ledger  ·  Production extract
Accounting-ops agent that closes the books over typed MCP tools. Auto-books only what a deterministic policy gate proves safe, routes the rest to a human, prints a full audit trail.

The Librarian  ·  Production extract
Agentic RAG over a podcast archive: hybrid BM25 + dense retrieval (RRF), answers cited to the minute, scored against a hand-labeled gold set (recall@k, MRR).

Different domains, one architecture: agents get typed, allowlisted tools with no raw access; a policy gate is enforced in code where a prompt cannot override it; state changes are reconciled against a source of truth; and anything that leaves the system passes a human. Building the same shape over many kinds of data is the point. The pattern and the discipline behind it are documented in Dev System.

LinkedIn · github.com/janvrsinsky