# Ali Cem Topcu AI Agent & MCP Developer AI Agent & MCP Developer specializing in context engineering: designing the information structures, memory systems and tool interfaces that determine what LLMs know, when they know it and how reliably they act on it. Enterprise infrastructure background applied to governed, phase-gated AI systems. Builds production MCP servers (shell-stream, txt-mcp, resume-json) and ONE TOOL, a private platform for governed agentic AI development, available for live demo. Also experienced in agentic-AI-assisted pipeline construction and in persistent AI memory systems, from multi-format context stores to vector-based semantic memory. Runs a boutique data recovery practice on a selective basis. Available for remote work globally and open to relocation worldwide. Location: Turkey (remote, open to relocation). Contact: contact@alicem.dev · LinkedIn: https://www.linkedin.com/in/ali-cem-topcu-b223a5a4 · GitHub: https://github.com/Kilgor ## Experience ### AI Agent & MCP Developer, Independent Development May 2025 – present · Remote (Turkey) Designing and building AI systems where context is treated as architecture: structured information delivery to LLMs via MCP servers, memory layers and governed agent pipelines. - Built 3 production MCP servers: shell-stream and txt-mcp published open source; resume-json as a private production tool - Engineered a multi-format memory architecture (MD/JSON/TXT) for persistent LLM context across sessions - Designed TourGuide, a JSON navigation system that controls what an AI reads and in what order - Built a phase-gated context delivery pipeline in ONE TOOL with human-approved context at each phase boundary - Built and deployed a fully offline local semantic memory pipeline (ChromaDB, BGE-M3, Ollama) controlled by a 33-tool FastMCP server Technologies: MCP, FastMCP, Python, AI Agents, Context Engineering, LLM Integration, FastAPI, ChromaDB, Ollama ### Freelance Data Analyst (project basis), SistemNow Jan 2025 – present · Istanbul, Turkey Project-based engagement for a logistics client's analytics infrastructure: maintaining and extending the Power BI and Power Query ETL pipeline, cleaning operational data and delivering dashboards. - Rebuilt a logistics company's data analysis infrastructure with a full ETL pipeline replacing spreadsheet workflows - Delivered interactive dashboards for real-time KPI monitoring Technologies: Power BI, Power Query, ETL, Data Modeling, Data Cleaning ### Freelance Data Analyst, Self-employed Jun 2024 – May 2025 · Turkey (remote) Transition year from hardware engineering into data and AI, building Python and data analysis skills through client projects. - Delivered Power BI dashboards for multiple clients - Built data extraction and preprocessing workflows with Pandas, NumPy and Matplotlib - Applied machine learning for predictive analysis and trend forecasting Technologies: Power BI, Python, Pandas, NumPy, Matplotlib, Machine Learning, EDA ### Data Recovery Engineer (selective cases), Independent Data Recovery Practice Jun 2024 – present · Turkey (remote / on-site) Continuing boutique data recovery practice on a selective basis alongside AI development, accepting complex cases that need specialized PC-3000 expertise. - Selective case acceptance with a high success rate Technologies: PC-3000 UDMA, Data Recovery Engineering, HDD/SSD/Flash Storage ### IT Support Manager, Online Telecom Mar 2022 – Sep 2022 · Turkey Short part-time engagement at a telecommunications group, alongside the data recovery practice: internal IT infrastructure, support workflows and daily operations. - Managed internal IT operations and support team - Maintained network infrastructure, servers and endpoint systems Technologies: Windows Server, Network Administration, IT Support ### Data Recovery Engineer, Independent Data Recovery Practice Mar 2018 – Jun 2024 · Turkey Full-time independent practice specializing in critical and complex storage failures, using PC-3000 UDMA professional tooling for hardware-level intervention. - PC-3000 UDMA certified by ACE Lab (Prague) - More than 95% customer satisfaction across hundreds of recovery operations - HDD, SSD and flash recovery from physical and logical damage - Enterprise-grade failures requiring low-level firmware and hardware work Technologies: PC-3000 UDMA, Firmware Analysis, Data Integrity, Storage Systems ### Technical Coordinator & Co-founder, Bilge İzmir Aug 2017 – Mar 2018 · Izmir, Turkey Co-founded an education and IT consulting firm and led its technical infrastructure work. - Designed and implemented network and server infrastructure - Technical lead for all IT delivery Technologies: Windows Server, Network Architecture, System Administration ### Founder & Systems Engineer, Infra Black (own company) Jan 2015 – Jan 2017 · Turkey Founded and operated an independent IT consulting company delivering infrastructure design, system administration and network engineering to private-sector clients. - Delivered end-to-end infrastructure solutions: server setup, network design, administration - Managed client engagements, technical delivery and operations as sole founder Technologies: Windows Server, Network Architecture, System Administration ### IT Infrastructure Consultant, Infopark Jan 2012 – Jul 2014 · Ankara, Turkey IT infrastructure consulting for public-sector and private-sector clients, including electricity distribution network localization and a critical government email system upgrade. - Led IT infrastructure design and implementation for UEDAŞ and ÇEDAŞ electricity distribution corporations - Delivered the MS Exchange 2007 to 2010 upgrade for the Turkish Prime Ministry - Designed server rooms, network architecture and Windows Server 2008 R2 systems - Managed backbone, firewall and routing configuration for mission-critical government systems Technologies: Windows Server, MS Exchange, Network Architecture, Infrastructure Design ## Projects ### ONE TOOL Governed platform for building AI agents - Role: Creator & Lead Developer - Period: Nov 2025 – present - Status: active - Domain: ai - Type: case-study - Human page: https://alicem.dev/projects/one-tool #### Summary A private, local-first platform for building agents under human-defined boundaries. Each phase controls exactly what the LLM receives, and every step is gated, logged and auditable. #### Case study ##### Problem Most agent-building platforms are cloud-based or cloud-dependent, and almost none are local and system-focused. I wanted AI to never operate without human-defined boundaries. ##### Approach A strict assembly line. A governed operation library catalogues domain functions under explicit rules. Those catalogued operations become the structured context for agent construction. Multi-format memory (MD, JSON, TXT) carries context across every phase boundary. ##### Architecture Phase-gated pipeline with human approval at each boundary, an operation library, memory layers, and a navigation layer (TourGuide) that tells the AI what to read. ##### Result Produces framework-agnostic agent templates (MCP, LangChain, OpenAI, CrewAI). The platform is built and maintained using itself. Note: Private production system. A live demo is available on request: contact@alicem.dev #### Highlights - Human-governed assembly line: AI is never unsupervised at any phase - Phase-gated agent generation with framework-agnostic templates - Multi-format memory architecture used across all phases - Fully local deployment, no external dependencies #### Technologies Python, FastMCP, MCP, AI Agents, Context Engineering, System Architecture ### Local Semantic Memory Pipeline Fully offline knowledge base for LLMs, controlled over MCP - Role: Designer & Developer - Period: May 2026 – present - Status: production - Domain: ai - Type: case-study - Human page: https://alicem.dev/projects/local-semantic-memory #### Summary A document indexing pipeline that gives LLMs fast, accurate access to curated project knowledge without sending any data to external APIs. #### Case study ##### Problem Onboarding LLMs need quick access to project-specific knowledge, and none of it should leave the machine. ##### Approach A four-stage pipeline: file discovery with change detection, cleaning and translation, structured parsing, and indexing into ChromaDB with BGE-M3 embeddings. Each document gets a metadata header before embedding so vectors are identity-aware. ##### Architecture Two independent FastAPI services (pipeline and logger) controlled by a FastMCP server with 33 tools across six groups: service management, pipeline control, indexer, health, configuration and log/query. ##### Result In production since v0.4.0 (September 2026), indexing five collections and queried by several LLM clients at once. The whole system runs through natural-language tool calls, with no manual terminal work. Note: Private production system. A live demo is available on request: contact@alicem.dev #### Highlights - Incremental processing with crash-safe checkpoints: stop and resume at any point - Rollback-safe production cutover with dated snapshot backups - Config-driven: one config file drives every component, updated live over MCP - Fully offline, no cloud dependencies #### Technologies Python, FastAPI, FastMCP, ChromaDB, BGE-M3, Ollama, Vector Databases, Semantic Search ### TourGuide JSON navigation guides that control what an AI reads - Role: Creator & Maintainer - Period: Dec 2025 – present - Status: active - Domain: ai - Type: case-study - Open source: yes - Source code: https://github.com/kilgor/TourGuide - Human page: https://alicem.dev/projects/tourguide #### Summary Human-authored, JSON-only guides that define exactly what an AI reads, in what order, and what to skip. It reduces hallucination caused by unstructured codebase exploration. #### Case study ##### Problem When you switch between AI assistants, each new one needs your codebase explained again. Unguided exploration of a large codebase is also where hallucinations start: the AI reads the wrong files, in the wrong order, and fills the gaps with guesses. ##### Approach A guide is a JSON file that you write once, with AI help. It says what to read, in what order, what to skip, and where the checkpoints are. TourGuide never discovers, generates or decides anything; the human is the cartographer. Setup takes 15 to 30 minutes for a small project and one to two hours for a large one: copy a template, explore the codebase with an AI, customize the guide, then test it on a fresh AI session. ##### Architecture A root tourguide.json links to sub-guides for deep dives, forming a map of interconnected JSON files. The repository ships templates, real examples, documentation and a demo you can show to an AI assistant. ##### Result Published open source under Apache 2.0. Used in production as the codebase navigation layer inside ONE TOOL, so every new AI session starts from the same complete context. #### Highlights - Apache 2.0 open source with templates and real-world examples - No code generation and no automation: pure context architecture - Used in production as the codebase navigation layer inside ONE TOOL #### Technologies Context Engineering, JSON, AI Agent Navigation, Open Source ### chunk-wizard Token-budget chunking library - Role: Creator & Solo Developer - Period: Jun 2026 - Status: released - Domain: ai - Type: case-study - Open source: yes - Source code: https://github.com/kilgor/chunk-wizard - Human page: https://alicem.dev/projects/chunk-wizard #### Summary A single-file Python utility that splits text into token-budget-aware chunks before it goes to an LLM. It detects the text's natural structure and picks the splitting strategy that keeps chunks as full as possible without exceeding the budget. #### Case study ##### Problem LLMs have a fixed context window. Feed them too much and they truncate silently or fail; split every N characters and you cut sentences in half and destroy structure. In my own pipelines, local models with deliberately limited context windows could not process the large documents going into a vector database. ##### Approach Measure first: the character budget comes from the model's context window, a share reserved for the reply, and an estimate of characters per token. Then detect the text's natural structure and try four strategies: Markdown headings, paragraphs, lines and sentences. Each produces a set of chunks scored by utilization, and the best-filled strategy wins. ##### Architecture One Python file with no dependencies. An in-memory function returns the chunks for a text; file-based functions split a file into numbered chunk files and reassemble them later; a third function only calculates chunk boundaries. Every operation returns the same result structure with success, data, error and metadata. ##### Result Released as open source under the MIT license and built for local LLM vector database pipelines. Its limits are documented: a single unit larger than the budget passes through whole, the token estimate is approximate, and only text files are supported. #### Highlights - Token budget calculated from the model's context window, not arbitrary sizes - Tries four strategies (heading, paragraph, line, sentence) and keeps the one with the best utilization - No dependencies (Python standard library only); MIT licensed #### Technologies Python, NLP, Local LLMs, Vector Databases ### shell-stream MCP Safe shell execution for AI assistants - Role: Solo Developer - Period: Jun 2025 – Aug 2025 - Status: released - Domain: ai - Type: open-source - Source code: https://github.com/Kilgor/shell-stream-mcp #### Summary An MCP server that lets AI assistants run and monitor shell commands with automatic security filtering, session management and real-time streamed output on Windows, Linux and macOS. #### Highlights - Filtering layer blocks destructive operations and allows safe administration - Audit logging of approved and blocked commands, plus a mode to test commands against the filters without running them - Session management with streamed output for long-running processes #### Technologies MCP, Python, FastMCP, Security Filtering, Process Management ### txt-mcp Text file read, write and edit over MCP - Role: Solo Developer - Period: Aug 2025 – Sep 2025 - Status: released - Domain: ai - Type: open-source - Source code: https://github.com/Kilgor/txt-mcp-server - HTTP variant: https://github.com/Kilgor/txt-mcp-server-http #### Summary A minimal MCP server that gives AI assistants three tools for UTF-8 text files: read, write, and search-and-replace edit. It works with any text-based file type, such as .txt, .json and .md. #### Highlights - Read, write and edit any text-based file over MCP - Built on FastMCP, with an MCP prompt that asks for confirmation before acting - Minimal footprint: three tools, one small file - A second variant adds a confirmation workflow for writes (confirmed flag) and SSE, HTTP and stdio transports #### Technologies MCP, Python, FastMCP, File I/O, Text Processing ### resume-json AI-managed resume with a single source of truth - Role: Developer & Architect - Period: Sep 2025 – present - Status: active - Domain: ai - Type: case-study - Human page: https://alicem.dev/projects/resume-json #### Summary A private MCP server that lets an AI assistant read, back up and safely update my structured JSON resume, and keep a log of job applications. It is the tool I use most actively, and the content source for this site. #### Case study ##### Problem A resume is a different document for every job, but the facts behind it should live in one place. I also wanted an AI assistant to help tailor it without ever being able to quietly overwrite or lose the original. ##### Approach One structured JSON schema (now at version 2.9) holds career history, skills, certifications, projects and hobbies. The server is a pure data provider: it makes no AI calls of its own. The AI client does all the analysis and tailoring, and the server only reads, writes and backs up. Anything that overwrites the resume needs explicit user confirmation, and a backup is made automatically first. ##### Architecture Three-layer clean architecture: a domain layer (models and operations), an application layer (the resume service) and an infrastructure layer (the FastMCP server and configuration). Seven tools. Read: read_resume, list_resume_backups, read_usage_log. Write: save_customized_resume and restore_resume_backup (both confirmation-gated), create_resume_backup, and add_usage_log_entry. Safeguards: file path validation against directory traversal, schema validation, and no API keys anywhere. ##### Result In active use since late 2025. About a dozen automatic snapshots between December 2025 and September 2026 trace how the resume evolved, and a job-application log records which version went to which role. It now also feeds this site, through a privacy filter that decides which fields may be published. Note: Private tool. A live demo is available on request: contact@alicem.dev #### Highlights - Pure data provider: no external AI calls and no API keys; the AI client does the analysis - Anything that overwrites the resume is confirmation-gated, with an automatic backup first - Seven tools covering the resume, its backups and a job-application log - Built with the ONE TOOL process: operation library extraction, phased and tested - Source of truth for this site, behind a privacy filter #### Technologies MCP, Python, FastMCP, JSON, Clean Architecture, LLM Integration ### Logistics Data Pipeline & BI System Analytics infrastructure rebuilt for a logistics company - Role: Data Engineer (Lead) - Period: Jan 2025 – present - Status: ongoing - Domain: data - Type: project #### Summary End-to-end data engineering project that rebuilt a logistics company's analytics infrastructure: ETL pipelines, data cleaning and modeling, and interactive Power BI dashboards for real-time KPI monitoring. #### Highlights - Replaced ad-hoc spreadsheet workflows with structured ETL pipelines - Full ownership: data audit, cleaning, modeling, dashboard design and client delivery #### Technologies Power BI, Power Query, ETL, Data Modeling, SQL, Data Cleaning ### Steam Gaming Industry Analytics 2024 Revenue and performance analysis of Steam games released in 2024 - Role: Data Analyst - Period: Sep 2024 - Status: completed - Domain: data - Type: project - Source code: https://github.com/kilgor/steam-2024-1500-games-by-revenue #### Summary Collected and plotted the top 1,500 Steam games released in 2024 by revenue (data gathered through September 2024). #### Technologies Python, Pandas, Matplotlib, EDA, Data Collection ## Skills ### Technical - Context Engineering (advanced, 1 yr) - MCP (Model Context Protocol) (advanced, 1 yr) - FastMCP (advanced, 1 yr) - System Architecture (advanced, 6 yr) - Data Recovery (PC-3000) (expert, 6 yr) - Problem Solving (expert) - Python (intermediate, 2 yr) - AI Agents (intermediate, 1 yr) - LLM Integration (intermediate, 1 yr) - AI Memory Systems (intermediate, 1 yr) - Agentic AI-Assisted Development (intermediate, 1 yr) - Local LLMs (intermediate) - Vector Databases (intermediate) - FastAPI (intermediate) - Machine Learning (intermediate, 1 yr) - Git/GitHub (intermediate, 3 yr) - Power BI (intermediate, 1 yr) - Microsoft Power Query (intermediate, 1 yr) - Microsoft SQL Server (intermediate, 1 yr) - Data Engineering (intermediate, 1 yr) - Data Analysis (intermediate, 1 yr) - Data Cleaning (intermediate, 1 yr) - Pandas (intermediate, 1 yr) - NumPy (intermediate, 1 yr) - Matplotlib (intermediate, 1 yr) ### Soft skills - Self-Directed Learning - Complex Problem Solving - System Design Thinking - Open Source Contribution - Technical Documentation - Adaptability ### Languages - Turkish (native) - English (advanced) ## Education - Bachelor of Science, Computer Engineering: Çankaya University (2001 – 2009) ## Certifications - Supervised Machine Learning: Regression and Classification: Stanford University (Coursera) (2024) - Unsupervised Learning, Recommenders and Reinforcement Learning: Stanford University (Coursera) (2024) - AI for Everyone: Stanford University (Coursera) (2024) - Python Programming and Data Analysis: An Bilgi (2023) - PC-3000 Data Recovery Professional: ACE Lab (2018) ## Interests - Open Source Contribution: Building and sharing MCP and AI tooling on GitHub. - AI & Technology Learning: Self-directed learning, currently focused on agentic AI patterns. - Scuba Diving: PADI diving instructor certification. - Self-Defense Training: Self-defense and martial arts. - Mountain Biking: Challenging trails and terrain. - Computer Gaming: Various genres.