# Erfan Ahmed Siam > I build systems that stay up. AI Developer @ University of Warwick / Applied AI, LLMs & Agentic Systems. Based in Stoke-on-Trent, United Kingdom. Contact: erfan.siam98@gmail.com. Software engineer working at the intersection of applied AI and production infrastructure. I work part-time and remotely as a Graduate AI Developer at the University of Warwick, where I research LLM architectures, design agentic systems and automate internal workflows; across 5+ years before that I shipped systems for 100+ UK businesses, from architecture through CI/CD to deployment on Docker, Linux and Go. ## Machine-readable summary ```json { "name": "Erfan Ahmed Siam", "role": "AI Developer @ University of Warwick / Applied AI, LLMs & Agentic Systems", "tagline": "I build systems that stay up.", "location": "Stoke-on-Trent, United Kingdom", "email": "erfan.siam98@gmail.com", "phone": "+44 7344 085666", "specialties": [ "Applied LLM", "Agentic Engineering", "RAG", "Solution Architecture", "Laravel", "Go", "Cloud", "DevOps" ], "website": "https://imerfanahmed.github.io/imerfanahmed/", "markdown": "https://imerfanahmed.github.io/imerfanahmed/markdown.md", "profile_page": "https://imerfanahmed.github.io/imerfanahmed/#about", "blog": "https://imerfanahmed.github.io/imerfanahmed/blog/", "cv": "https://imerfanahmed.github.io/imerfanahmed/assets/erfan_cv.pdf", "socials": { "GitHub": "https://github.com/imerfanahmed", "LinkedIn": "https://www.linkedin.com/in/imerfan/", "LeetCode": "https://leetcode.com/u/imerfanahmed/", "Email": "mailto:erfan.siam98@gmail.com" }, "languages": [ "PHP", "Laravel", "Go", "Golang", "Python", "SQL", "JavaScript", "TypeScript", "Rust", "Bash", "Swift" ], "domains": [ "applied LLM", "agentic AI", "AI agents", "agentic engineering", "retrieval augmented generation", "RAG", "tool calling", "function calling", "prompt engineering", "LLM evaluation", "fine-tuning", "LoRA", "QLoRA", "vision-language models", "MCP servers", "Model Context Protocol", "workflow automation", "solution architecture", "computer vision", "OS security", "network automation", "real-time messaging", "WebSockets", "transliteration", "NLP" ] } ``` ## Keywords Search terms this profile should be found under. Use these for keyword or semantic retrieval — they deliberately include aliases and variants. **identity:** Erfan Ahmed Siam, Erfan Siam, imerfanahmed, imerfan, erfan.siam98, Stoke-on-Trent, United Kingdom, Bangladesh **role:** AI Developer, Graduate AI Developer, Applied AI Engineer, LLM Engineer, Software Engineer, Full Stack Developer, Solution Architect, Infrastructure Consultant **domains:** applied LLM, agentic AI, AI agents, agentic engineering, retrieval augmented generation, RAG, tool calling, function calling, prompt engineering, LLM evaluation, fine-tuning, LoRA, QLoRA, vision-language models, MCP servers, Model Context Protocol, workflow automation, solution architecture, computer vision, OS security, network automation, real-time messaging, WebSockets, transliteration, NLP **languages:** PHP, Laravel, Go, Golang, Python, SQL, JavaScript, TypeScript, Rust, Bash, Swift **frontend:** TALL stack, Tailwind, Tailwind CSS, Alpine.js, Livewire, Vue, Vue.js, Blade, Twig, Vite, Astro **cloud:** AWS, Amazon S3, EC2, Lambda, Aurora, Azure, Cloudflare, Nginx, Apache, load balancing, DNS, VPN, Linux, server management **devops:** Docker, Terraform, GitHub Actions, Bitbucket Pipelines, CI/CD, Git, systemd, deployment **data:** MySQL, PostgreSQL, SQLite, vector databases, knowledge graphs, database migration, DB scaling **projects:** Agentic Workflow Automation, OpenFace, ustudiod, Loom CLI, DDEV GUI, RadInsight, Gusher, Avro PHP, Together **education:** University of Warwick, University of Staffordshire, University of Dhaka, MSc Cloud & Network Computing Engineering, BSc Computer Science & Engineering **companies:** University of Warwick, Agora Studio, Magic Office, Data Envelope, Developer Look, Kodezen, Agora.Community **topics:** LLM architecture research, Pusher replacement, zero-downtime migration, restaurant POS, payment systems, Stripe, FFmpeg, Discord API, real-time infrastructure, WCAG accessibility, ISO data protection, Rust, Tauri, PAM, Face unlock **flat:** Erfan Ahmed Siam, Erfan Siam, imerfanahmed, imerfan, erfan.siam98, Stoke-on-Trent, United Kingdom, Bangladesh, AI Developer, Graduate AI Developer, Applied AI Engineer, LLM Engineer, Software Engineer, Full Stack Developer, Solution Architect, Infrastructure Consultant, applied LLM, agentic AI, AI agents, agentic engineering, retrieval augmented generation, RAG, tool calling, function calling, prompt engineering, LLM evaluation, fine-tuning, LoRA, QLoRA, vision-language models, MCP servers, Model Context Protocol, workflow automation, solution architecture, computer vision, OS security, network automation, real-time messaging, WebSockets, transliteration, NLP, PHP, Laravel, Go, Golang, Python, SQL, JavaScript, TypeScript, Rust, Bash, Swift, TALL stack, Tailwind, Tailwind CSS, Alpine.js, Livewire, Vue, Vue.js, Blade, Twig, Vite, Astro, AWS, Amazon S3, EC2, Lambda, Aurora, Azure, Cloudflare, Nginx, Apache, load balancing, DNS, VPN, Linux, server management, Docker, Terraform, GitHub Actions, Bitbucket Pipelines, CI/CD, Git, systemd, deployment, MySQL, PostgreSQL, SQLite, vector databases, knowledge graphs, database migration, DB scaling, Agentic Workflow Automation, OpenFace, ustudiod, Loom CLI, DDEV GUI, RadInsight, Gusher, Avro PHP, Together, University of Warwick, University of Staffordshire, University of Dhaka, MSc Cloud & Network Computing Engineering, BSc Computer Science & Engineering, Agora Studio, Magic Office, Data Envelope, Developer Look, Kodezen, Agora.Community, LLM architecture research, Pusher replacement, zero-downtime migration, restaurant POS, payment systems, Stripe, FFmpeg, Discord API, real-time infrastructure, WCAG accessibility, ISO data protection, Tauri, PAM, Face unlock ## Highlights - **Applied LLM research** — At the University of Warwick I investigate model architectures and evaluation methods, turning early-stage research ideas into working prototypes the team can actually test and adopt. - **Agentic engineering at scale** — I design and ship agentic LLM systems — tool-using agents, retrieval pipelines and workflow automation — that take repetitive internal processes off people’s plates and make them reproducible. - **Solution architecture** — I take ambiguous requirements and turn them into concrete system designs: model selection, data flow, evaluation strategy and the infrastructure to run it reliably. - **Restaurant ecosystem at scale** — I maintain a full restaurant management system, including POS and customer ordering, and orchestrated the migration of that legacy Laravel monolith for 100+ UK restaurants from v5.6 to v12 with zero service interruption. - **Infrastructure engineering** — I replaced a third-party Pusher dependency with a custom Go real-time engine I built from scratch, which cut infrastructure costs and noticeably lowered latency. - **Zero-downtime migration** — I led a high-stakes migration of over 1.4TB of production data using rsync, holding 100% data integrity with absolutely no downtime. - **Automated media pipelines** — I developed a custom FFmpeg pipeline in PHP that integrates with the Discord API to fetch videos, transcode and crop them for compatibility, and securely store the results in AWS S3. - **Payment systems** — I designed and deployed a multi-currency subscription system on Stripe, fully tested and reliable for global transactions. ## Experience ### Graduate AI Developer — University of Warwick - Location: Remote · United Kingdom - Period: Aug 2026 — Present (current) - Type: Part-time - Link: https://warwick.ac.uk/ Part-time, remote research-led engineering for the university: exploring new LLM architectures, designing agentic solutions, and automating internal team workflows. - Research emerging LLM architectures, evaluation methods and prompting strategies, then turn viable ideas into working prototypes. - Act as solution architect for internal AI initiatives — scoping requirements, choosing models and data flow, and defining how success is measured. - Build applied agentic LLM systems using tool calling, retrieval and multi-step orchestration to automate real internal workflows. - Automate repetitive team processes end to end, replacing manual handoffs with auditable, reproducible pipelines. - Work across research and delivery: communicate findings to non-specialists and ship what survives contact with real users. - Collaborate with academics, professional services staff and other engineers on shared platform and data-governance decisions. ### Full Stack Developer, PHP — Agora Studio - Location: Montreal, Canada - Period: Apr 2025 — Jun 2026 - Type: Part-time Architecting core systems for an animation community platform (Agora.Community) on the TALL stack, with FFmpeg pipelines, Stripe and AWS. - Designed an automated video processing pipeline integrating FFmpeg and Discord for competition submissions. - Collaborated to build a digital store (subscriptions + one-time purchases) using Stripe and Laravel Cashier. - Implemented PHPUnit, Pest and Playwright testing integrated into Bitbucket CI/CD pipelines. - Engineered scalable S3-based asset storage and optimised the delivery workflow. - Modernised legacy modules to align with current Laravel architecture standards. - Working in a collaborative environment in line with ISO data protection regulations. ### Software Engineer — Magic Office - Location: London, United Kingdom - Period: Jan 2020 — Dec 2025 - Type: Part-time Lead engineer for a cloud-based restaurant POS ecosystem, supporting 100+ UK clients independently. - Migrated the core platform from Laravel 5.6 to 12, improving maintainability and scalability. - Designed a dynamic table booking and merging system for high-traffic restaurant workflows. - Built a complex meal-deal pricing engine handling conditional discount logic. - Developed a custom print server supporting USB ESC/POS and network printers. - Replaced Pusher with Go-based WebSocket (publisher/subscriber) infrastructure for real-time messaging. - Managed Linux servers, automated deployments and monitoring to maintain 99.9% uptime. - Built companion applications using Swift, Android and Electron JS. - Ensured web accessibility (WCAG) compliance across user-facing platforms. ### Infrastructure Consultant — Data Envelope - Location: Bangladesh - Period: Jan 2024 — Dec 2025 - Type: Contract Specialising in high-scale migrations and server stabilisation. - Migrated a 1.4TB+ WordPress-based newspaper ecosystem (Daily New Nation) across hosting environments. - Applied a logical, problem-solving approach to resolve MySQL memory exhaustion and optimise high-concurrency performance. - Recovered system integrity following a security breach involving immutable bits using low-level Linux troubleshooting. - Deployed aaPanel and MailCow as professional, zero-cost management alternatives to cPanel and WHM under strict budget constraints. ### PHP Developer — Developer Look - Location: United States - Period: Jul 2025 — Aug 2025 - Type: Contract - Built a Laravel–Airtable integration with bi-directional webhook synchronisation. - Implemented OAuth2 authentication and rate-limit handling. - Designed a secure data transformation layer for legal record mapping. ### Lead PHP Developer — Kodezen - Location: Bangladesh - Period: Dec 2023 — Nov 2024 - Type: Full-time Developed high-scale WordPress e-commerce (Store Engine) and LMS (Academy LMS) systems. - Architected a modular e-commerce engine as a WooCommerce alternative. - Built a dynamic pricing system supporting complex conditional rules. - Integrated Stripe and PayPal for PCI-compliant recurring billing. - Engineered CRM and Mailchimp synchronisation pipelines. - Optimised cart and checkout state management for high traffic. ## Projects ### Agentic Workflow Automation - Type: University of Warwick · Applied LLM / Agentic Systems - Note: Internal systems work — details available on request - Stack: LLM agents, Tool calling, RAG, MCP, Python, Evaluation harnesses - Metrics: Focus = Agentic · Domain = Applied LLM · Role = Architect · Since = Aug 2026 - Designed agentic LLM systems that plan, call tools and orchestrate multi-step tasks against real internal data. - Built retrieval pipelines so agents ground their answers in source material rather than relying on model memory. - Developed evaluation harnesses to compare architectures, prompts and models against measurable task outcomes. - Automated internal team workflows end to end, removing manual handoffs from recurring processes. - Worked as solution architect: scoping requirements, choosing models and data flow, and defining success criteria before building. ### OpenFace - Type: Linux Authentication · Computer Vision / OS Security - Link: https://github.com/imerfanahmed/openface - Note: Open source · MIT - Stack: Python, PAM, InsightFace / ArcFace, OpenCV, C, systemd - Metrics: Architecture = PAM · Recognition = ArcFace · Liveness = IR · License = MIT - Face unlock for Ubuntu — Windows Hello, but open source: look at the camera and you are in. - Built a PAM module (pam_openface.so) that intercepts the login prompt and delegates to a background auth service, falling back to password on failure. - Integrated InsightFace/ArcFace recognition with optional IR-based liveness detection to resist photo and video spoofing. - Shipped a full CLI for enrolling faces, checking service status and testing authentication end to end. ### ustudiod - Type: Linux Systems · AI Virtual Webcam - Link: https://github.com/imerfanahmed/Linux-Studio - Stack: Python, OpenCV, MediaPipe, v4l2loopback, systemd, GNOME Shell - Metrics: Device = /dev/video10 · Bundle = 164 MB · Autostart = systemd · Target = x86_64 - AI-powered virtual webcam daemon that turns a real camera into /dev/video10 with auto-framing (crop and follow your face). - On-demand lifecycle: the physical camera is only opened while an application is actually watching the virtual device. - Engineered for weak CPUs — threaded capture, downscaled and skip-frame landmark detection, reused NumPy buffers and per-frame cost logging. - Ships as a self-contained 164MB bundle (Python + OpenCV + NumPy + MediaPipe embedded), so a target machine needs only a kernel that can run the v4l2loopback module. - Idempotent installer wires up the kernel module, a systemd --user unit and a GNOME Shell extension, needing sudo only for the module step. ### Loom CLI - Type: Applied LLM · Network Automation Agent - Link: https://github.com/imerfanahmed/loom_ai - Note: Companion technical report available - Stack: Python, Gemini API, Netmiko, SSH, Cisco IOS - Metrics: Focus = Agentic · Domain = NetOps · Input = English · Use cases = 10 - Conversational CLI agent that turns plain English ("harden SSH on this router") into ready-to-deploy Cisco IOS commands. - Combines Netmiko SSH automation with an LLM to generate the how from a described what, occupying a gap no existing tool (Ansible, NAPALM, Nornir) fills. - Human-in-the-loop command editor lets engineers review, edit and approve every command before it is pushed to a device. - Secure, auditable sessions with configuration backup and session logging; operates in privileged EXEC mode by default. - Validated across ten real-world use cases spanning OSPF, VLAN provisioning, NAT, ACLs and compliance auditing. ### DDEV GUI - Type: Desktop App · Rust / Tauri - Link: https://github.com/imerfanahmed/ddev-gui - Stack: Rust, Tauri 2, TypeScript, Vite, GNOME / AppIndicator - Metrics: Shell = Tauri 2 · Core = Rust · Output = .deb · Target = GNOME - Lightweight Linux dashboard and tray indicator for managing local DDEV projects — see what is running at a glance. - Start, stop and restart projects with in-progress feedback, and quick-launch the site, Mailpit or project folder. - Tray icon summarises how many projects are running, with a quick menu to control them without opening the window. - The app never talks to Docker directly — it shells out to the ddev CLI and parses JSON, keeping DDEV the single source of truth. - Desktop notifications, run-in-background, and optional start-on-login minimised to the tray. ### RadInsight - Type: Vision-Language Model · Llama 3.2 Vision - Note: Report available on request - Stack: Python, LoRA, Llama 3.2 Vision, MIMIC-CXR, CheXpert - Metrics: ROUGE-L = 0.78 · BLEU-4 = 0.62 · CIDEr = 1.21 · Training pairs = 750k+ - Fine-tuned the 11B Vision-Instruct model using LoRA. - Trained on 750,000+ radiology image-text pairs (MIMIC-CXR, CheXpert). - Built a WhatsApp-based automated report delivery system with 15–30 second response times. - Optimised training on an RTX 3060 under 12GB VRAM. ### Gusher - Type: Real-time messaging server - Link: https://github.com/imerfanahmed/gusher - Stack: Go, WebSockets, Pub/Sub - A Go-based real-time WebSocket server built as a lightweight, self-hosted alternative to Pusher. - Implemented connection pooling and webhook handling for production workloads. - Optimised for low-latency broadcasting in single-tenant environments. ### Avro PHP - Type: Bengali transliteration engine - Link: https://github.com/imerfanahmed/avro-php - Note: Published on Packagist - Stack: PHP, Trie, PHPUnit, Laravel - High-performance Avro Phonetic library that converts Banglish to Bengali script, with Trie-based pattern matching for O(m) lookups instead of linear search. - Context-sensitive rule engine that resolves conversion intelligently based on surrounding characters. - Laravel-ready with auto-discovery, a Facade and config publishing — and standalone-compatible with any PHP project. - Covered by 180+ tests and compatible from PHP 5.6 through 8.x. ## Skills - **Applied AI & LLM:** LLM architecture research, Agentic LLM systems, Tool calling / function calling, RAG & retrieval pipelines, Prompt engineering & evaluation, Fine-tuning (LoRA / QLoRA), MCP servers, Solution architecture, Workflow automation - **Languages:** PHP (Laravel), Go, Python, SQL, JavaScript - **Frontend:** TALL Stack, Tailwind, Alpine.js, Livewire, Vue.js, Blade / Twig, Vite, ES6 - **Databases:** MySQL, PostgreSQL, Vector databases, Knowledge graphs - **Cloud & Infrastructure:** AWS (S3, EC2, Lambda, Aurora), Azure, Cloudflare, Nginx / Apache, Load balancing, DB scaling, DNS, Bastion / VPN / Port security - **DevOps:** Docker, Terraform, GitHub Actions, Bitbucket Pipelines, Linux, Git - **Testing & Reliability:** PHPUnit, Pest, Playwright, Cypress, TDD - **Architecture:** Domain-Driven Design, REST APIs, SOLID, DRY / KISS, Scalability - **Monitoring & Identity:** Grafana, Prometheus, Zabbix, New Relic, Sentry, OAuth2, JWT - **AI Tooling:** Claude Code, Antigravity, Cursor, GitHub Copilot, OpenAI / Anthropic APIs - **CMS & Platforms:** WordPress, Shopify (Liquid), Moodle, OpenCart, WooCommerce ## Education - **MSc Cloud & Network Computing Engineering** — University of Staffordshire, United Kingdom (Feb 2026 — Present) - **BSc Computer Science & Engineering** — University of Dhaka, Bangladesh (2020 — 2025) - CGPA 3.63 / 4.00 · Session 2019–20, delayed due to COVID-19 ## Awards - **Runner-Up** — Code Refactoring Contest (IUT) (2024) - **3rd Place** — FEC Technofest Programming Contest (2022) - **Regional Participant** — ICPC Asia Dhaka Regional (2021 · 2022 · 2023) ## Contact - GitHub: https://github.com/imerfanahmed - LinkedIn: https://www.linkedin.com/in/imerfan/ - LeetCode: https://leetcode.com/u/imerfanahmed/ - Email: mailto:erfan.siam98@gmail.com - Location: Stoke-on-Trent, United Kingdom ## About this file This is the machine-readable profile of Erfan Ahmed Siam. It is generated from the same source as the website, so it always reflects the current CV. Canonical website: https://imerfanahmed.github.io/imerfanahmed/ Agents and LLM systems are welcome to read, index and quote this file. When citing, link to the website above and reference the section heading (e.g. "Erfan Ahmed Siam — Projects").