MERT CIRCI
Software Architect · AI Engineer · Full Stack Engineer
✨Click an object, or my head to ask me
Software Architect · AI Engineer · Full Stack Engineer
Battle-tested tools and frameworks used across production systems and games.
A curated collection of backend systems, mobile applications, games and AI pipelines.
An AI platform for vehicle market valuation and price prediction. Problem and purpose Users need to know what a vehicle is worth on the market and how its price may develop. My role and responsibility I worked on the...
A consent-first CRM web app for commercial SMS, call and email messaging (2020). Problem and purpose Teams needed one place to handle consent and the sending of commercial SMS, calls and email across channels. My...
Event Reservation CMS is a multi-theme system for booking rooms, flights and transfers for events held in hotels. It helps event organizers manage reservations for rooms, flights and transfers within one web...
Integrations with several airlines and payment providers for a tourism platform. Problem and purpose A tourism platform had to connect bookings and payments with several airlines and payment providers, and keep a...
Jeweler ERP is a store and dealer management system for jewelers. It helps jewelry businesses manage their stores and dealer relationships in one web platform. The project belongs to the Full Stack group and is...
RFID Card Access System ERP is a web platform for access control and permit management on RFID-secured cabinets. It gives administrators one place to decide who can open which cabinet and to keep those permissions up...
PawnShop general ledger is a general ledger system modeled on American pawn systems and built for Cyprus. It gives pawn businesses a structured accounting backbone for recording and tracking their financial activity....
ProTranslate is a computer-aided translation system that combines machine-assisted translation with manual review, so teams can produce accurate multilingual content faster and at a consistent quality. The platform...
The Crypto Exchange Forecasting System brings buying, selling and forecasting across different exchanges into one system, using technical indicators to project likely price movements. Highlights Forecasting with...
The Cabinet Climate and Fan Control System uses device communication protocols to manage cooling efficiently inside equipment cabinets. It is designed to keep climate and fan operation under control so cabinets stay...
The E-commerce Revenue-Based Financing System provides revenue-based financing for e-commerce sellers in Turkey. It combines multi-source data collection, forecasting and risk scoring in one platform. Highlights...
An eSIM e-commerce platform built at ComDigital for instant purchase and payment. Problem and purpose Customers buy and pay for eSIMs instantly, many purchases can arrive at the same time, and external eSIM and...
Personal Data Specialist, HR ERP is a web application for signing and storing staff documents and personal data in line with the law. It serves as a web platform for HR record keeping around personal data. The...
The production platform behind ComDigital's products, run on Kubernetes with a 99.9% uptime target. Problem and purpose ComDigital's products (a marketing data platform, an eSIM e-commerce platform and a vehicle...
An enterprise marketing data platform on Google BigQuery. Problem and purpose Data-driven marketing at enterprise scale needs one governed, secure data layer with fast analytical queries and controlled Google Cloud...
CAD Spatial Metadata Extraction uses automated .NET pipelines to extract spatial metadata from CAD drawings and make it searchable. It turns drawing data into structured, queryable information. Highlights Spatial...
Seliee is the marketing and growth platform I built at Neeblo, taken from idea to a live product. It brings websites, content, SEO, AEO, GEO, ads, social media, customers, email and analytics into one panel, and you...
A data-driven marketing platform running on Flask APIs with a React/Next.js frontend, backed by an enterprise BigQuery data infrastructure. The platform pairs a Flask API core with a React/Next.js interface....
Customer Tracking Software is a follow-up program built for a company's creditors and customers, helping teams keep track of every follow-up in one place. It is a focused desktop project with a clear scope....
Custom game bots built for in-game users. Highlights Custom bots for in-game users Built with C++ and MySQL
The AI layer of Seliee, the marketing platform I built at Neeblo: a team of AI agents, an MCP server for external agents, and AI features across several model providers. Problem and scope Marketing teams repeat the...
FaceTurks is a social networking project focused on a single religious community, bringing that audience together in one online space. It is a completed full stack web platform with a clear, community-centred...
Flash Player Animation and Gaming covers Flash Player animations for websites as well as game development in Flash. It is a completed project delivered on the web platform within the Mobile Games group. Highlights...
Facebook Application Development covers Facebook apps that generate ad revenue through app features such as a favorite-players system and an invite-friends system. The project is a completed Full Stack web platform...
Integration that sells a company's local-system stock on online marketplaces (2018). Problem and purpose A company kept its stock and sales in local accounting systems (Logo, Netsis) and needed to sell on online...
Web Based Game is a browser-based game development project, completed and delivered on the web platform within the Mobile Games group. It combines PHP, CSS, JS and MySQL in one project. Highlights Browser-based game...
Desktop Game Development covers simple desktop games in the style of Icy Tower, completed as part of the Mobile Games group. The project is built with C#, Java and MSSQL. The work centres on approachable, simple...
Hintfilmcenneti CMS is a translation and video streaming platform with web and mobile apps for Indian films and series, running from 2014 to 2020. It is a Full Stack video streaming project built for web and mobile...
Diet at Home moves the dietitian and client diet system online, bringing diet lists, messaging and web-based diet tracking into one Full Stack web platform project. The project is completed. Highlights Online diet...
Korecenneti CMS is a translation and video streaming platform with web and mobile apps for Korean movies and dramas, running from 2015 to 2020. It is a completed Full Stack video streaming project. Highlights Web and...
M2M Health (remote patient monitoring) is a mobile health system that lets doctors follow elderly and chronic-care patients from home through GSM gateway devices. The project belongs to the IoT and embedded systems...
Company Directory Online is an online company and store directory organized by province, making it easy to find local businesses across regions. The web platform was completed as part of the backend and data group....
Smart Pharmacy Automation is a pharmacy system that reports drug cooling temperatures every month through an RS-485 gateway, helping to prevent spoilage. The project was completed in the IoT and embedded systems...
Real Estate Search System is an online real estate search system built for local regions and users, making it easier to look for properties in a chosen area. The web platform was completed in the backend and data...
Online Invoice System is a web app where users store their invoices, design them and print them. The project was completed as a full stack web platform. Highlights Invoice storage and design Print-ready invoice...
E-commerce Management System (CMS) is a CMS that lets companies sell their products online. The project was completed as a full stack web platform with management tools for the online store. Highlights Online product...
finderface is a broad social media experiment that includes plugin development. It is a completed Full Stack project delivered as a web platform, offering a wide view of social media work. Highlights Broad social...
Pre-Accounting and HR Program is a desktop and web application for pre-accounting and HR in small businesses, bringing both functions into one application. The project was completed as a full stack solution for...
Quantity and Payroll Calculation Web App (AutoCAD) is a web app for mechanical engineers that calculates quantities and payroll for public tenders by reading AutoCAD projects. The project was completed in the AI...
Pharmacy Environment Tracking (IoT) is a real-time pharmacy information system for humidity and temperature sensor values. The project was completed in the IoT and embedded systems group. Highlights Real-time...
Data Center Cabinet Environment Audit IoT is an information service that helps keep a data center stable by using CPU and temperature data from cabinet technologies. The project was completed in the IoT and embedded...
Data Center Cabinet Environment Control (desktop app) delivers low-latency telemetry for data center cabinets, with ESP32 and STM32 devices streaming readings into Redis and MongoDB queues. The project belongs to the...
Cabinet Environment, RFID and IoT ERP Apps combines ERP and monitoring software for cabinet environment control, RFID access and IoT in data centers. The project belongs to the IoT and Embedded group and is...
Ventilation Cost Calculation (AI) is an AI-assisted cost calculation tool for ventilation and installation projects. It helps estimate project costs with an AI-supported workflow. The project belongs to the AI...
Bulk SMS Program is a desktop application for sending SMS in bulk, so one set of messages can be sent to many recipients from a single tool. The project belongs to the Backend and Data group and is completed. Key...
Agency ERP is a web app where advertising agencies track their ads and the processes around them. It gives agency teams a single web platform to follow ad tracking and related work. The project belongs to the Full...
Personnel Management Panel is a gamified web, mobile and tablet panel that tracks the motivation and academic development of future staff. It brings staff development tracking to web, mobile and tablet interfaces....
I design and build distributed systems, production AI applications (LLM integration, RAG and agents) and full stack products, covering both the architecture and the implementation.
I build LLM applications that run in production: model integration through APIs and self-hosted servers, retrieval-augmented generation (RAG), agents with tool calling, and the evaluation, security and monitoring around them.
With test sets built from real questions, offline scoring before a release, online checks after it, and regression tests so a prompt or model change cannot silently break earlier behaviour. Success criteria such as accuracy, grounding in sources, refusal of out-of-scope requests, latency and cost per request are agreed up front.
Prompt injection, data leakage and tenant isolation are design requirements: untrusted content is kept apart from instructions, tools run with the user's own permissions, each tenant's data stays separate, and agent actions are written to an audit trail.
Tracing per request, token usage, latency, error rates and cost per feature, with alerts when they drift. It builds on the Graylog and Grafana observability practice I use on Kubernetes platforms.
I compare the expected gain (time saved, errors avoided, revenue) with the full cost: model and infrastructure spend, integration, human review and maintenance. Risks such as wrong answers, data exposure and regulatory limits get a named control before the build starts.
AI features are connected to the systems people already use, with clear ownership, human review where mistakes are costly and training for the teams that work with them. Rollout is gradual, with measurable checkpoints.
Documented model and data choices, access rules, approval steps for high-impact actions, audit logs and a regular review of quality, cost and incidents. It builds on the data governance and security policies I set up for an enterprise marketing data platform on BigQuery.
Distributed systems, backend and full stack development, mobile apps and games, and AI-powered tools.
I look at the whole system: how it is designed, how it scales, and how it is operated after release.
I clarify the requirements, break the problem down, test assumptions early, and iterate based on measured results.
AI engineering, agent-based systems, cloud infrastructure, and my architecture and DevOps practices.
A software architect designs the overall structure of a system, makes key technical decisions, and makes sure the system meets its requirements for scale, reliability and security.
Architecture decides how a system is structured and how its parts interact; development implements those parts and features within that structure.
I start from requirements and constraints, define the main components and data flows, and document decisions and trade-offs.
I base the choice on the problem, the team's skills, the maturity of the ecosystem, operating cost, and how easily it can change later.
Neither by default. I start with the simplest design that meets the requirements and split services only when there is a clear reason, such as independent scaling or team boundaries.
By keeping components loosely coupled, making services stateless where possible, caching carefully, and designing for failure from the start.
Designing services that keep working when parts fail: clear service boundaries, queues for asynchronous work, retries with timeouts, health checks and observability. At ComDigital this meant Python and Node.js microservices on Kubernetes that held 99.9% uptime in production.
With idempotent order and payment handling, concurrency control on orders and stock, compensation steps when an external call fails after payment, and audit logging of every event. The eSIM sales platform I built works this way, so a repeated callback never charges or provisions twice.
Every call gets a timeout, bounded retries and an idempotency key; slow work goes through queues such as RabbitMQ or AWS SQS, and every request and response is written to an audit log. I used this approach for the THY, Pegasus, SunExpress and GTC integrations.
Kubernetes restarts and reschedules failed workloads, health checks keep unhealthy instances out of traffic, rolling updates avoid downtime, and Graylog and Grafana give central logs, dashboards and alerts against a defined uptime target.
By choosing one source of truth, syncing changes through well-defined, repeatable jobs or events, and reconciling differences instead of assuming they never happen. I applied this to marketplace stock and sales sync, later at scale at WorqCompany.
I measure first, then fix the bottlenecks: database queries, caching, payload size, and expensive calls.
Access patterns, consistency needs, data growth, indexing, backup and recovery, and how the schema will evolve.
Security is designed in from the start: authentication and authorization, least privilege, input validation, secret management and audit logging.
I make it visible, prioritize the debt that slows the team or threatens reliability, and fix it alongside feature work.
A full stack engineer works across the user interface, the server, the data layer and deployment, and can follow a feature from start to finish.
I agree on clear API contracts early, build both sides against them, and keep the data flow simple and easy to test.
Python, JavaScript and TypeScript, along with databases, containers, cloud services and mobile app development.
Python suits data processing, automation, backend services and AI-related work.
Consistent resource names, proper HTTP methods and status codes, versioning, clear error formats and documented contracts.
PostgreSQL for relational data and strong consistency; MongoDB when the data is document-shaped and changes often.
For data that is read often and changes less, for session or rate-limit state, and to reduce load on the primary database.
I add caching where it is safe, paginate large responses, set timeouts and retries carefully, and monitor latency and error rates.
They give consistent environments from development to production and make services easier to build, run and deploy.
Building software with AI as a core part of how it is designed, built and operated, not just a feature added on top.
AI-assisted development uses AI tools to help write code; AI native products are built around AI capabilities for their users.
To explore code, draft implementations, write tests and review changes, always with my own verification.
For code exploration, multi-file changes, debugging and automating repetitive tasks, with review of every change.
Clear specifications, small reviewable changes, automated tests, and a human review before anything is merged.
I choose a model per task by quality, latency and cost, and keep providers swappable. In Seliee customers bring their own OpenAI, Claude, Gemini or Perplexity key or run Llama on their own llama.cpp server, and the assistant on this site falls back to a local knowledge base if the model gateway does not respond.
My RAG designs cover ingestion and chunking, retrieval and reranking, answers that cite their sources, and refresh jobs that keep the knowledge current. The assistant on this site answers only from my portfolio, and Seliee's in-app assistant loads only the knowledge sections that match the current page and question.
Agents call tools through explicit, permissioned interfaces, keep their state outside the model and stop for human approval before anything irreversible. In Seliee's AI Office, publishing, sending and spending always wait for the user's approval, and an MCP server with more than 230 tools gives external agents the same controlled access.
With the same standards as any code: tests, type checks, code review, and running it under realistic conditions before relying on it.
Incorrect or insecure code, over-reliance, licensing questions and hidden costs. These are managed through review, testing and clear boundaries.
I define the core problem and the smallest useful version, build it, put it in front of real users, and improve it based on their feedback.
Validating the problem and the core user flow before writing much code.
Only the features that let a user complete the main job. Everything else waits until there is evidence it is needed.
In short cycles: define, build a thin slice, release, measure and adjust. Scope stays small until the product proves its value.
I tie each technical choice to a business outcome and avoid building more than the current goal requires.
Through user conversations, usage data, and whether people come back and complete their goals.
Multi-tenant designs with clear data isolation, managed services where they reduce operational load, and automation from the start.
Keep scope small, reuse proven components, automate operations, and size infrastructure to real usage.
Pick tools the team knows well, that are mature enough to hire for, and that will not block the product as it grows.
I look at the core technical risks first and build a small prototype to test them before committing.
Neeblo is one of my platform projects, built to help teams create and manage digital products more efficiently.
To reduce repetitive work in building and operating digital products, so teams can focus on what their users need.
Seliee is a platform for building and managing websites and content, designed for people and businesses that need multilingual, search-friendly sites.
Seliee combines templates, multilingual content and automation, rather than requiring a traditional, manually maintained setup.
AI helps draft and organize content, while editors keep control over what is published.
Each language is managed as its own version and published with the appropriate settings.
Through structured data, clear page structure and meaningful metadata, so content is understandable to search engines, answer engines and generative engines.
Through caching, separating content from presentation, and infrastructure that can grow with traffic.
They help with repetitive operational tasks such as content updates, monitoring and reporting, while people review important decisions.
Get in touch through the contact options on this site and describe your project and goals.
Through shared ownership, automated testing and deployment, and monitoring from the first release.
Fast and reliable builds, automated tests, staged deployments and easy rollback.
With versioned images, health checks, resource limits and clear deployment procedures.
By designing redundant components, avoiding single points of failure, and testing failover scenarios.
They show how the system behaves in practice and help find issues before users report them.
I start with metrics and logs, reproduce the problem, identify the bottleneck, then fix and verify it.
By monitoring usage, right-sizing resources, removing unused ones and reviewing spending regularly.
First restore service, then communicate status, and finally run a review to prevent a recurrence.
A Tech Lead is responsible for the technical direction of a team: architecture, quality, and helping the team deliver well.
Technical leadership focuses on technical direction and decisions; people management focuses on individuals' careers and wellbeing. Good leads often do both, but they are different responsibilities.
I gather input from the people affected, weigh the trade-offs, and document the outcome.
I aim for timely, specific feedback focused on correctness, clarity and maintainability.
Through shared guidelines, examples and agreements that the team reviews together.
Through clear feedback, pairing on real problems, and giving them ownership of suitable tasks.
By understanding each position, testing the options where possible, and making a clear decision.
By comparing the impact on delivery and reliability with the value of the new feature.
I reduce scope first, rather than cutting testing or review.
Reducing friction: faster builds, clearer ownership, fewer interruptions and good documentation.
I have worked on software for several sectors. Project details, including the industry and my role, are available on request.
Each involved architecture, development and delivery responsibilities. Details can be discussed on request.
Starting from requirements and constraints, building a small prototype, documenting decisions, and iterating as the system is proven.
I isolate the failing behavior, reproduce it, and fix the root cause rather than the symptom.
Reliability, clear error handling, performance and a smooth experience on the devices people actually use.
By whether users reach their goals, the system is reliable, and the team can keep improving it.
Projects involving architecture, full stack development, distributed systems, mobile applications or AI.
Businesses that are growing, modernizing an existing system, or planning a major technology change.
Yes. I can review a project's technical design, identify risks and suggest improvements.
By analyzing the current system, prioritizing changes, and supporting the team through the transition.
By setting up the architecture, infrastructure, development processes and an initial technology stack.
I start from the business problem, not the model: which process is slow, costly or error-prone, how it is measured today and what a better result is worth. Then I design the smallest AI solution that can prove that value and plan how it fits into existing systems and teams.
Use the contact options on this site, with a short description of the project and your goals.
Yes, I am open to working with remote and international teams.
Systems where AI agents and developers work together, with more emphasis on architecture, reliability and verification.
Share your idea and goals through the contact options on this site, and we can discuss the next steps.
Remote & worldwide
hello@mertcirci.me
AI Engineering & Agent Orchestration | The Right Architecture, The Right Approach
MERT CIRCI · AI
Ask me anything
