Portfolio — 2026
Alexander
Snurnitsyn
I work across machine learning, quant trading, and game design — and I write and self-publish fiction. Most of it solo, end to end.
Games · Machine Learning · Writing · Trading
Selected work ↓The full history — roles, education, and skills — on one page.
View CV →Selected work
Algorithmic Trading System
A LightGBM cascade turning years of market data into a long-only book — with a monitoring console, and an obsession with not fooling myself.
View →Space RTS
A real-time 3D fleet-combat prototype — Total War in space, with EVE-inspired ship handling.
View →
Image LoRA + ComfyUI
Experiments in ComfyUI: training a style model for a particular quality of light, and the pipelines around it.
View →Fiction Books
Two fantasy novels — written, translated from Russian, self-published, and marketed end to end.
View →Systems & Data
Larger engineering builds.
Algorithmic Trading System
A LightGBM cascade turning years of market data into a long-only book — with a monitoring console, and an obsession with not fooling myself.
The biggest thing I've built: a long-only equity strategy driven by a cascade of LightGBM models (with a fine-tuned FinBERT in the stack) trained to predict return over SPY across ~1000 names, a portfolio and risk layer tested across many configurations, and a full web console for monitoring and control. It runs end to end on paper trading, deployed on Hetzner. Built entirely solo. I keep it to architecture and method — no published results or edge claims until it's properly validated.
Trading Signal Platform
A deployed stock-signal platform — one analysis engine behind a web app, a Telegram bot, and portfolio/risk tools.
A stock-signal platform I built and run: MACD/RSI/Volume signals with KNN forecasting and AI news analysis, plus portfolio, risk, and trading-journal tools. One analysis engine feeds three surfaces — a web app (primary, deployed on Hetzner, multi-user), a Telegram bot, and an older desktop client. Nothing exotic under the hood; the value is that it's a real, shipped, full-stack product.
Neuroevolution Sandbox
A NEAT-style artificial-life world — creatures with evolving neural-network brains, no fitness function, evolution you watch in real time.
Project ALife: a population of creatures lives in a 2D world, each driven by its own recurrent neural network. There's no fitness function and no training loop — they eat, move, reproduce, and die, and over generations their genomes and brains mutate (brains can grow new neurons). You watch evolution happen live and can click any creature to inspect the network inside its head. Early and experimental, but a genuinely fun one — GPU-batched with PyTorch and streamed to a PixiJS frontend.
AI & Tools
Models, pipelines, and creative tooling.
Image LoRA + ComfyUI
Experiments in ComfyUI: training a style model for a particular quality of light, and the pipelines around it.
Self-directed image-generation work in ComfyUI. I was after a particular quality of light rather than a cartoon style, so I built a pipeline to generate and curate training data (IP-Adapter + ControlNet from a reference), trained a style model on kohya_ss — a plain LoRA underfit the look, so I moved to a LoHA with the Prodigy optimizer — and experimented with multi-stage generation (style, upscaling, face/hand restoration) across Flux, SDXL, and Wan 2.2.
StoryForge — Text-Style LoRA
A controlled experiment: does structured tagging make a fine-tuned LLM more steerable?
A data-centric fine-tuning project. I built a pipeline to select, dedup, and diversify a corpus of interactive-fiction scenes, tag them against a hand-curated closed vocabulary with an LLM, and train two QLoRA adapters — one on plain scenes, one on the same scenes conditioned on control tags — to test whether tagging buys controllable stylistic generation. The interesting part isn't the fine-tuning; it's the data and the experiment design.
StoryForge — Book Editor
An AI editing pipeline for book-length fiction — chapter by chapter, with context that respects the timeline.
A tool I built for editing my own manuscripts: it edits a book chapter by chapter while carrying evolving context (a running summary, versioned character profiles, an author style profile), and — crucially — when editing chapter N it only sees what's known up to N−1, so nothing from later chapters leaks backward. Edits come back as inline annotations, not silent rewrites, keeping a human in the loop.
Games
Design docs and playable prototypes.
Space RTS
A real-time 3D fleet-combat prototype — Total War in space, with EVE-inspired ship handling.
A solo Godot prototype for real-time tactical fleet battles: data-driven ships, a tactical order system, and a deterministic tick-based combat simulation with object pooling and LOD. It reached a sandbox where you command dozens of ships in real time — before I shelved it, ahead of UI, missions, and enemy AI.
Pandamonium
A manual-input action roguelike with a morality-driven Balance system — my first Godot project.
An action-roguelike prototype: a panda with a bamboo staff where every attack is a manual click, not an auto-battler. My first game project, built systems-first — a data-driven ability system with combos, an item system, a morality-based Balance mechanic that buffs and debuffs stats, a between-waves shop and economy, and enemy wave spawning.
GearTopia
A Factorio-like factory sim with Dwarf-Fortress depth levels and no electricity — everything runs on gears and shafts.
A factory-building game in Godot (C# + GDScript): dig down through Dwarf-Fortress-style depth levels and power your factory entirely through mechanical energy — shafts, gears, and gear ratios instead of electricity. A custom simulation of RPM/torque networks, data-driven buildings, conveyors, and cross-level item transfer, with the heavy systems written in C# for performance. My most technically ambitious game; currently paused.
Storage — Game Design Document
The hideout's loot-management module: a unified stash, drag-and-drop, and furniture-driven capacity.
A full GDD for the Storage module of Mortalis — the hideout hub where players manage everything they extract from the arena. Covers a unified inventory/stash interface, category tabs, filtering and sorting, drag-and-drop rules, and a system where craftable furniture drives both the room's look and its storage capacity.
Workbench — Game Design Document
The hideout's crafting station, built around recipe knowledge you earn rather than unlock.
A full GDD for the Workbench module of Mortalis — the hideout's production station. Covers the crafting workflow, a shared inventory/storage resource pool, and a recipe-research system where crafting knowledge is discovered and earned rather than handed out.
Background
A BBA at Riga Business School (with an exchange semester in South Korea) and a run of applied, consulting-style projects — market analysis, fintech regulation, marketing growth. Before that, a stint as a junior accountant at Civitta, where the real work turned out to be automating the process around the job — a VBA macro that cut one recurring task from three hours to thirty minutes.
Along the way, a research project comparing self- versus traditionally- published books across the top 100 Amazon Kindle — which I later put to use publishing and marketing my own novels.