Alihan Ibrayev

Full-stack software engineer and applied AI specialist with 3+ years building production AI systems in an academic medical environment. Designed and deployed "Distributor" — a custom LLM orchestration platform powering exam grading, document summarization, semantic search, and multimodal content processing on on-premise GPU hardware.

Founded

MenuRaskladka, a SaaS daycare operations product. Skilled in RAG pipelines, vector databases, multi-agent architectures, LLM inference tuning, and agentic development workflows.

Full-Stack Software Engineer at University of Alberta (2022-06 – Present)

Designed and built a medical student management platform for the MD Program, supporting user management, elective scheduling, reporting, and communications for hundreds of students annually.

• Delivered full-stack features using Python, FastAPI, Laravel, Node.js, and Vue; implemented infrastructure and deployment pipelines with Nginx, Docker, and Git.
• Built an email bus, Google Workspace integrations, and reporting tools that automated previously manual administrative processes.

Research & Development Lead (AI Systems) at University of Alberta (2022-06 – Present)

Designed and built "Distributor" — a custom LLM orchestration platform with task queuing, retry logic, parallel execution, and persistent WebSocket connections to on-premise DGX GPU servers, with SQL-backed run history and per-job rerun capability.

• Engineered per-task context window allocation and memory isolation strategies to maximize parallel GPU throughput across concurrent LLM workloads of varying complexity.
• Built an AI-powered exam grading system using local Gemma models with multi-level grading skills (strict, lenient, partial-credit) and multimodal image description. Achieved Quadratic κ = 0.864 and Pearson R = 0.857 vs. human graders, with LLM-generated reasoning traces for instructor audit.
• Built a multi-skill summarization service: an OSCE skill consolidating examiner comments into structured feedback, and an MSPR skill producing polished student performance summaries used in residency applications.
• Built a hybrid semantic tagging pipeline using Qwen embeddings and Neo4j vector search with LLM reranking and confidence scoring across a large curated tag taxonomy.
• Built XtoText — a media extraction service (PDF, Word, images, audio, video) using OCR and Nvidia Parakeet speech-to-text, with chunked embeddings powering RAG-based calendar search with permission-level metadata filtering.

Software Engineer at Nematic

Built a production grading platform using a hybrid computer vision and LLM pipeline that outperformed GPT-4 and Claude 3 on bubble-sheet recognition by using purpose-built CV over general vision models.

• Engineered a virtual bubble-sheet overlay system with OpenCV and Google OCR landmark detection, providing robust fill detection on smeared or poorly scanned documents.
• Built an OCR + LLM intake pipeline for respondent identification, automated handwritten response extraction, and open-ended answer analysis.

Founder & Full-Stack Developer at MenuRaskladka (Present)

Founded and built a SaaS daycare operations platform for meal planning, nutritional tracking, inventory management, invoicing, and financial reporting, with full PDF document export.

• Implemented an AI-assisted menu generator: users configure nutritional targets and meal preferences; system generates optimized weekly meal plans automatically.
• Building a teacher and parent portal: daily lesson plans with auto-pulled meal data, student activity logging with media, and parent-facing check-in/check-out with digital sign-off.

Founding Engineer / Full-Stack Developer at Kidventures Digital Inc.

Built a children's activity marketplace with search, recommendations, scheduling, booking, and payment processing, designed for multi-provider scalability.

B.Sc. in Double Major in Physics and Mathematics – University of Alberta (2021)