Projects
Selected work spanning research and production: the proof-of-learning and watermarking code behind my papers, real-time systems primitives, interactive explainers, and clinical tooling. One through-line runs through all of it: ML and systems you can verify, not just run. Each project links to its repository and writeup; the Interactive Lab has eight playable demos of the ideas.
Avion AI Strategy & Agent Platform
Authored Avion's enterprise AI strategy and built the platform that presents it. A three-tier architecture (air-gapped on-prem LLMs, a governed cloud-LLM API tier, and edge) plus a suite of retrieval-augmented and autonomous agents: an engineering-knowledge copilot with grounded citations and change-impact analysis, an autonomous repository agent, an RFP accelerator, and a training-debrief writer. Built end to end with Flask, SQLite, Docker, and GitLab CI/CD. Explore the interactive reference architecture →
Research Lab: interactive experiments
Eight interactive experiments covering distributed consensus, model watermarking, proof-of-learning, fault tolerance, and gradient descent. Each presents a real scenario with measurable outcomes. Built to make the core concepts of my research accessible in three minutes.
Research
SecurePoL with Watermarking
Reference implementation accompanying my dissertation. Integrates feature-based model watermarking with proof-of-learning to defend ML training integrity against spoofing attacks.
Adversarial Examples for Proof-of-Learning
Code investigating adversarial strategies that subvert PoL verification. Informs defenses introduced in the SecurePoL line of work.
METU Master's Thesis
Detection of cyber-security events from Turkish Twitter and newspaper streams, completed at Middle East Technical University (METU) under Prof. Cengiz Acartürk. This research served as the foundation for my published ICISSP paper, co-authored with him.
Applied AI Concepts & Architectures
A collection of brainstormed enterprise AI architectures and proof-of-concept system designs. These models explore how discipline-specific AI agents, grounded in approved corporate data, can utilize local on-prem inference for data sovereignty alongside secure APIs for de-identified reasoning.
-
Engineering Knowledge Copilot
A version-aware copilot over a massive engineering corpus. Answers questions with grounded citations, synthesizes the impact of document and spec changes across downstream artifacts, and generates suggested playbooks. Powered by on-prem multimodal parsing and semantic diffs.
-
Autonomous Engineering Staff Agent
A hybrid agent embedded in the software development lifecycle. Conducts autonomous code reviews, verifies release integrity against requirements, and flags integration risks before deployment using self-hosted Git pipelines.
-
RFP Accelerator
An agent that ingests Request for Proposal (RFP) documents, searches past answers, product specs, and OEM manuals, and drafts a clear, cited reply. Reduces RFP cycle-time significantly without data exfiltration.
-
Instructor Debrief Writer
Enhances training session logs by stripping student identifiers, drafting a polished on-tone debrief from telemetry and instructor bullet notes, and suggesting rubric grades for human approval.
-
Customer Update Auto-Writer
Cuts incident turnaround time by translating technical fault summaries into customer-ready bulletins that are on-brand, non-alarmist, and free of proprietary detail, utilizing a PII redaction gateway.
-
Fault Diagnostics Copilot
Retrieves similar historical cases and OEM bulletins via RAG to classify the likely fault domain (software / hardware / wiring) and assign ownership on the first pass using an air-gapped on-prem model.
-
Brand-Controlled Visual Generation
A localized Stable Diffusion XL / ControlNet pipeline that enforces strict style-locking and automated brand-compliance checks, generating high-fidelity assets in seconds.
-
Qualification Lifecycle Copilot
Generates hardware qualification plans with time estimates, tracks checklist completion live, and auto-packages test results and telemetry into audit-ready bundles with full requirement traceability.
-
Leadership Briefing Generator
Produces weekly leadership pulses and meeting agendas from a single source of truth by reading commits, pull requests, and support threads through a redaction gateway before summarization.
-
Predictive Maintenance Forecaster
Uses fleet-wide operational telemetry to predict component wear and schedule maintenance proactively before a system goes unplanned-offline. Built on Scikit-learn and Prophet time-series models.
Production Systems (Industry)
Real-time platforms for Level D flight simulators
Architecting real-time simulation platforms and cloud-native infrastructure for Level D full-flight simulators: the highest certification level, where simulator time counts as flight time.
Avion Control Engine (ACE) Architecture Report
A massive production-grade Instructor Operating Station (IOS) and Telemetry Dashboard built to command A320 flight simulators. Powered by a SvelteKit frontend and Python/gRPC backend. Read the architectural deep-dive (PDF).
Havelsan DLP: data-leakage prevention at national scale
Led the 14-engineer team that delivered a data-leakage-prevention product for defence and government clients.
Mission & ground control for Kargu & Togan UAVs
Developed critical mission-control and ground-control systems for Kargu, Turkey's first indigenous loitering-munition (kamikaze) UAV and an internationally recognized autonomous system, and the Togan reconnaissance UAV. Hard-real-time, safety-critical C++.
Secure Web Gateway · Patch Manager · Dragon browser
Led design and architecture of Comodo's Secure Web Gateway, enterprise Patch Manager, and the Chromium-based Dragon browser.
Applied Tooling
EKG Image Measurement Tool
Specialized desktop utility for cardiology professionals. Built and field-tested with clinicians to provide reliable, real-world scale calibration for EKG diagnostics.