Simple and cost-effective
I'm not a fan of over-engineering. I believe in keeping things simple, fast, and cost-effective. My goal is to build architectures that get out of the way and let the hardware—and the team—do their best work.
About
Huntay AzerTürk is a software architect with 13+ years of experience. This page covers my approach, selected projects, technologies, experience and writing.
About
Depth beyond the résumé
Applied AI — proven delivery
I've built autonomous support systems using Meta WhatsApp, Chatwoot, and Dialogflow CX. By integrating RAG architectures, we improved help desk workload and response times by 30–70% and customer satisfaction by 15–40% across different projects.
Security-first architecture
Security isn't an afterthought for me—it's the foundation. I build systems that don't just scale when things go right, but hold strong when things go wrong.
Academic lens
Cognitive science is my research focus and a field I keep studying and reading about—it gives me a distinct angle on software. I don't just see AI as a tool; I see it as a way to design systems that align with how humans actually think and work.
Writing & research
I regularly write on Medium about things like RAG chunking, nature-inspired networks, and the reality of AGI. In my spare time I also work independently on Brain-Computer Interfaces (BCI) using EEG signals and LLMs for neurodivergent individuals.
Discipline as a system
I bring the same discipline from my CrossFit and boxing routines into my code. High performance requires focus, consistency, and a strong foundation—whether in the gym or in system architecture.
Projects
Selected work: the problem, what was done and the outcome.
Bare Metal: 4x throughput on a single-core Celeron
.NET · Multi-threading · Async · Hardware
- Action
- Inherited a centrifuge/spectrometer control system that crashed on i7 + 8GB RAM. Stripped out the bloat, made the mechanical and optical pipelines fully async, and pulled analysis to the core.
- Outcome
- 4x throughput while dropping the hardware requirement to a single-core Celeron with 1GB RAM. Real Bare Metal performance.
Hexagonal: 60+ service cloud → 3 Windows services
Hexagonal · Ports & Adapters · Monolith · On-premises
- Action
- Closed/military sites couldn't touch the internet. Because the system was built around Hexagonal Architecture and clean interfaces, I bypassed the API Gateway and merged just the Application Layer into a monolith.
- Outcome
- Kubernetes and RabbitMQ swapped for SQLite, MQTT, and in-memory. The full 60+ service cloud platform now runs as 3 lightweight Windows services—without changing a single line of business logic.
AI-supported help desk system
Help desk · RAG · LangGraph · Dialogflow CX
- Action
- Designed a support system based on LangChain, LangGraph and RAG that answers repetitive fault and documentation questions from dealers, technicians and users using manuals and past tickets; integrated it on GCP with Dialogflow CX, Chatwoot and WhatsApp.
- Outcome
- Across different projects, help desk workload and response times improved by 30–70% and customer satisfaction by 15–40%, with a 10–30% improvement in costs and sales.
Reports without waiting for a developer
Reporting · ERP data · Self-service
- Action
- Every new report needed a developer. We built an engine that lets users create, filter and schedule complex reports themselves.
- Outcome
- Teams create and schedule their own reports; the developer bottleneck for reporting is gone.
10,000+ field devices on one platform
Field devices · IoT · Mobile · Service data
- Action
- Field devices of 8 types (fire panels, intruder alarms, cameras, suppression systems) were managed through an unreliable outsourced system. We replaced it with a single platform of our own, with web and mobile apps.
- Outcome
- One platform now serves 10,000+ devices and 100,000+ users, and engineers can debug a single faulty device without disrupting the rest.
Production sandbox without disrupting 10,000 devices
Production debugging · A/B routing · Zero downtime
- Action
- Designed an A/B-test-style routing layer that lets engineers redirect a single faulty device or user into a sandbox/debug environment—live, on production traffic.
- Outcome
- Faster root-cause analysis with zero downtime for the other 10,000+ devices in production.
Technical stack
Cloud, AI, and leadership—where I operate.
Core Engineering & Languages
- .NET / C#
- Python
- C / C++ (Hardware/Legacy)
- Java (Low-level/Android)
- TypeScript / React / Blazor
- SQL / NoSQL
AI & Cognitive Science
- LangChain & LangGraph
- RAG & Vector Databases
- LLM Orchestration
- Computer Vision (YOLOv8)
- Dialogflow CX
- AI Agents in ERP
Cloud & Distributed Systems
- Kubernetes & Docker
- GCP (Google Cloud)
- RabbitMQ & MQTT
- Terraform
- Microservices
Architecture & Leadership
- Hexagonal Architecture
- Bare Metal Optimization
- Event-driven Design
- High-availability (Zero-downtime)
- CEH / Security-first approach
- Engineering Management
Experience
Dec 2025 – Present
Software & Digital Solutions Manager
Teknim Elektronik San. Tic. A.Ş.
Istanbul, Turkey · Hybrid
Leading the entire IoT and backend infrastructure for fire safety operations—focused on scalability, reliability, and opening up new markets.
Architected an 'Agentic' support system using LangChain and LangGraph for complex chain-of-thought reasoning, looping, and branching, powered by Python vector databases (RAG).
Later integrated this autonomous architecture into GCP with Dialogflow CX, Chatwoot, and Meta WhatsApp, improving help desk workload and response times by 30–70% across different projects.
Future projection: migrating ERP data to BigQuery for AI agent swarms.
Ongoing · spare time
Brain-Computer Interfaces (BCI)
Independent
I work on this on my own, in my spare time: Brain-Computer Interfaces (BCI) using EEG signals and LLMs for neurodivergent individuals.
Nov 2020 – Nov 2025
Software Architect & Team Lead
Teknim Elektronik San. Tic. A.Ş.
Istanbul, Turkey
Replaced an outsourced and unreliable infrastructure by building a 60-70 microservice ecosystem from scratch on Linux—Docker, Kubernetes, RabbitMQ, MQTT.
Single platform serving 8 different device types (fire panels, intruder alarms, cameras, suppression systems) for 10,000+ devices and 100,000+ end users.
Migrated legacy monoliths to event-driven microservices on GCP—roughly 75% infrastructure cost reduction.
Designed a Hexagonal Architecture that lets the same business logic run as either a 60+ service cloud platform OR as 3 lightweight Windows services for closed/military networks—zero business code change.
Built a one-click sandbox/debug architecture so any faulty device or user could be redirected to a safe environment without disrupting the 10,000+ devices in production.
2022
Software Architect — Dynamic Systems
Enmos
Istanbul, Turkey
Designed a meta-programming layer: a runtime SQL query and reporting engine using .NET, Blazor, React, DevExpress, and NoSQL.
Eliminated developer dependency for new reports by implementing dynamic filtering and scheduled cron-jobs for enterprise data analysis.
Oct 2018 – Oct 2020
Senior Software Engineer
Teknim Elektronik San. Tic. A.Ş.
Istanbul, Turkey
Replaced an unmaintainable legacy C/C++ fire alarm configuration tool with a clean .NET core in just 45 days using Agile and TDD.
Industry-first: removed physical license dongles and shipped an ERP-integrated autonomous licensing service—then launched the first web-based fire monitoring system using React + MQTT.
Built a mini-MES system that automatically reads device serial numbers on the production line and flashes the correct firmware—running for 6 years with zero production errors.
Refactored critical C++ components into maintainable .NET Core microservices with Redis and RabbitMQ for high-throughput hardware messaging.
2016 – 2018
Senior Backend Developer & Team Lead
Nano Mabna Iranian
Iran
Inherited a centrifuge and spectrometer control system that crashed even on i7 + 8GB RAM. Refactored it from the ground up with async pipelines.
True Bare Metal win: 4x throughput increase while dropping the hardware requirement to a single-core Celeron with 1GB RAM.
Wrote custom algorithms turning raw optical signals into clinical lab results.
2013 – 2016
Software Engineer — Real-Time Critical Care
Pooyandegan Rah Saadat
Iran
Built a zero-latency monitoring system for life-critical hardware: 14-channel ECG, SpO2, ventilators—using pure multi-threading and ConcurrentQueue to fully decouple I/O from UI.
Synchronized data streams over both TCP/IP and Serial Port for bedside and central multi-patient monitoring.
Early IoT vision: streamed ambulance data directly into doctors' screens as an early warning system.
2013
Full-Stack Developer — Clinical Automation
Independent Project
Iran
Built a complete electronic health record (EHR), appointment, and finance management system for clinics on .NET + SQL Server.
2011
Mobile Security Engineer — Digital Forensics
Law Enforcement Cyber Crime Project
Iran
Developed an Android (Java) application for authorized mobile forensics and evidence collection, built for a law enforcement cyber crime unit.
Hands-on work with Android internals, permission models and vulnerabilities shaped the zero-trust approach I bring to security architecture today.
2009
Algorithm Engineer — Heuristic Optimization
Academic Research
Iran
Replaced brute-force search algorithms with custom Genetic Algorithms and Ant Colony Optimization—my first real foothold in applied AI.
Set the foundation for the resource-minimization mindset I still apply to every system I build.
Education
2012 – 2015
B.Sc., Computer and Information Systems Security / Information Assurance
University of Applied Science and Technology
Certifications & interests
- —CEH (Certified Ethical Hacker)
- —Software Architectural Design
Active CrossFit athlete—high-intensity training and performance discipline.
Selected writing
Articles and posts (LinkedIn and Medium).
The Cybersecurity Crisis in SMEs: The National Economy's Soft Underbelly in the Age of AI (in Turkish)
LinkedIn
Is AGI Really Possible?
Medium
The Weakest Link in RAG: Advanced Document Chunking
Medium
The Software World and AI: Revolution or Evolution?
Medium
Nature-Inspired Neural Networks (graph-based connectivity, mini memory)
Medium