01
AGENTIC SAFETY + MEMORY
LLMCORTEX
Local-first pre-reasoning tripwires for coding agents with deterministic matching, composable risk signals, audit logs and 219 passing tests.
Inspect repositoryOpen to remote AI/ML Lead roles
Applied AI for systems where correctness matters.
I turn ML and LLM prototypes into reliable products—from data and evaluation to architecture, deployment, observability, risk controls and iteration.
- 12+ yearsproduction engineering
- Agents + RAGgrounding, evaluation, guardrails
- Financial + edge MLPyTorch, ONNX, Core ML
Review the proof
Three public systems. Inspectable code and explicit trade-offs.
Agentic safety, financial ML and on-device document intelligence—not a gallery of toy prompts.
02
FINANCIAL ML + VALIDATION
GoldGRU
XAU/USD time-series research with BiGRU and self-attention, 18 selected features, walk-forward validation and explicit controls against overfitting.
Inspect repository03
LOCAL-FIRST DOCUMENT AI
EvidencePipelineKit
PDFKit and Apple Vision OCR with confidence, provenance, deterministic retrieval, contradiction gates, citations and explicit human-review outcomes.
Inspect repositoryWhat I own
From model behavior to production delivery.
Agentic systems and grounded RAG
Tool use, retrieval, citations, provenance, contradiction gates, retries, validation and explicit human review.
Evaluation and reliability
Offline and online metrics, leakage-resistant validation, reproducible tests, observability and honest uncertainty.
Financial ML and risk systems
Microstructure features, time-series models, realistic execution, fail-closed controls and operational risk limits.
On-device and cross-platform AI
Core ML/ONNX, Vision/PDFKit, OCR, Swift/Rust integration and private offline workflows that fail honestly.
Selected systems, described honestly
Evidence beyond demos.
01
Production agentic and RAG workflows
Grounded retrieval, source provenance, contradiction gates, retries, auditability and explicit human-review boundaries.
02
On-device AI + Rust platform
SwiftUI product backed by a 40-module Rust core, 209 unit tests, 14 integration tests, Core ML/ONNX delivery and reproducible Apple builds.
03
Camera PPG and real-time DSP
AVFoundation capture, signal-quality gates, stabilization, hysteresis, device validation and an honest non-medical product boundary.
04
Philosophy of Artificial Intelligence
Master’s candidate at Ufa University of Science and Technology, 2026–2028: AI methodology, philosophy of mind, knowledge engineering and societal risk.
Remote · Limassol, Cyprus