Skip to content
ET

Emir T

@emir-t

AI/ML Engineer & Simulation Developer

Applied AI, geospatial analytics, and simulation engineer with an electrical engineering background from ČVUT FEL (Prague).

About

Experienced in local VLM fine-tuning (QLoRA / NF4), deterministic industrial plant simulation, fault-tolerant ML backends, and multi-node agentic development workflows.

Career

Education

Electrical Engineering, Power Engineering and Management

ČVUT FEL, Czech Technical University in Prague

2023 — 2026

B.Sc., Electrical and Power Engineering

Satbayev University (Kazakh National Research Technical University)

2026

Selected projects

AURA, AI-assisted diagnostics and recovery for Docker applications

Built an operations assistant that investigates failures using logs, resource measurements and configuration history. Meta Muse Spark Open 1026 agents via Baseten propose typed recovery actions within administrator-defined policies. Verified Linkding recovery on a test setup in 93.3 seconds, with 3 model calls and 18 verification observations. Confirmed availability using real HTTP requests. Bu...

Open project

A.S.T.R.A.

Built A.S.T.R.A. solo and outperformed 145 teams and more than 500 participants. Urban environmental analytics and ML-based greening optimization. Architected a fault-tolerant geospatial backend modeling 40 urban micro-zones in Astana (Dockerized PostgreSQL + Redis and zero-config local SQLite WAL) with automated Open-Meteo / CAMS Global ingestion and multi-tier fallback to MET Norway API and l...

Open project

DriveIndUI, Automotive Plant Simulator

Built a deterministic 8-hour shift manufacturing simulator (250 ms tick rate, up to 60x speed) enforcing strict mass-balance invariants across Onix, Cobalt, and J7 lines, paired with a Three.js 3D shop-floor view and Fastify API. Connected a server-side risk predictor via OpenAI Responses API (GPT-6.1 Sol) for 15/30/60-minute horizons; isolated the model from future scenario data leaks and benc...

Open project

Local Bank Cash-Desk Video Compliance System

For a hackathon project, fine-tuned Qwen3-VL-8B-Instruct in 4-bit NF4 precision via QLoRA to run local visual audits of banking cash-desk footage without sending sensitive video streams to external cloud providers. Implemented a 3-second interval chronological video decomposition pipeline with automated visual verification to log staff compliance violations across full transaction timelines. Da...

Open project

Multi-Node Agentic Workflow & Custom Diffusion LoRAs

Configured a multi-node development environment with SSH access and parallel execution of development tasks, synthetic ML benchmarks, automated tests and project builds. Trained multiple custom LoRA checkpoints for Stable Diffusion: dataset preparation, rank/alpha tuning, step balancing, and weight validation.

Achievements

1st place, GreenHack Kazakhstan, A.S.T.R.A., solo developer

GreenHack Kazakhstan · 2026

Participant, Datathon ’26, local bank cash-desk video audit

Datathon ’26 · 2026

Participant, Qostanai Industry Hackathon, DriveIndUI automotive plant simulator

Qostanai Industry Hackathon · 2026

Participant, Indie Dev Track, AURA

Meta AI Developer Hackathon · 2026

Skills

LightGBMSHAP (TreeExplainer)Qwen3-VL-8B-InstructQLoRA (4-bit NF4)PEFTStable Diffusion (custom LoRA training)OpenAI Responses API (GPT-6.1 Sol)Meta Muse Spark Open 1026 / Basetenweight quantizationlocal inferenceFastAPIPostgreSQL (Neon)RedisSQLite (WAL)Docker ComposeWebSocketGoogle Cloud Compute EngineAstra 6 (Codex)Antigravity CLI (Argon / Opus / Flash 3.8)multi-node SSH workflows