From vague AI ideas to buildable systems
LLM applications, agentic workflows, RAG pipelines, structured extraction, evaluation criteria, human review, and traceable delivery paths.
Applied AI Engineer / AI Systems Architect
Alex Malinovsky, Applied AI Engineer.
I turn ambiguous AI ideas into maintainable systems: architecture, data flow, prompts, APIs, evaluation, review, deployment, and interfaces people can actually use. My background combines 10+ years in commercial software engineering with 4 years in regulated clinical-trials and medical-imaging R&D.
Operating Range
The useful overlap: AI architecture, regulated data, product interfaces, and production engineering.
Medical imaging, CV, and regulated R&D
4 years in Medical Trials and Medical Imaging, plus implemented AI/CV systems for real-time video stream analysis and crowd safety workflows.
Architecture that survives real usage
Frontend, backend, DevOps, MLOps, API boundaries, deployment paths, data-heavy interfaces, and shared platform code with quality gates.
Applied AI Engineer
AI Systems Architect
LLM / RAG Engineer
AI Product Engineer
Medical Imaging AI teams
AI-focused full-stack roles
Selected Proof Points
Text-first case studies until public demos and project media can be attached.
Agentic Solution Factory
Reusable multi-agent toolkit that turns raw client requirements into discovery, architecture, security, DevOps, estimation, review, and delivery packages.
On-prem RAG architecture
Designed a support automation architecture with vLLM, Qdrant, BGE-M3 embeddings, reranking, FastAPI, RBAC metadata filters, audit logs, and guardrails.
Real-time video safety system
Designed and implemented an AI/CV system for video stream analysis, crowd safety monitoring, risk detection, alerting, and operator-facing workflows.
Medical imaging platform work
Owned shared frontend architecture, libraries, testing standards, and production applications in a clinical-trials and medical-imaging R&D environment.
Engineering Bias
I am most useful when AI has to become a system, not a prompt demo.
Ambiguous requirements need architecture first
I turn vague AI goals into data flow, APIs, runtime boundaries, evaluation criteria, and delivery slices that teams can actually build.
Human review is a product feature
For clinical, support, and enterprise workflows, the system needs traceability, auditability, role boundaries, and clear failure behavior.
AI work still needs software engineering discipline
Prompts are only one layer. Maintainable AI products also need frontend ergonomics, backend contracts, tests, observability, deployment, and operational ownership.
Selected Experience
Senior individual contributor background across regulated R&D, frontend platforms, AI architecture, and automation.
Independent R&D / AI Product Engineering
- Built Agentic Solution Factory, a reusable multi-agent toolkit for discovery, architecture, security, DevOps, estimation, review, and delivery packages.
- Designed on-prem enterprise RAG architecture using vLLM, Qdrant, BGE-M3 embeddings, reranking, FastAPI, RBAC filters, audit logs, and prompt-injection guardrails.
- Implemented LLM-based resume parsing flows for PDF/DOCX documents with structured candidate extraction and reviewable outputs.
- Designed and implemented an AI/CV system for real-time video stream analysis and crowd safety management at large-scale events.
Clario / 4Create
- Worked as a senior individual contributor in a regulated clinical-trials and medical-imaging R&D environment.
- Owned maintenance and roadmap of 10+ shared libraries and 2 production applications used across multiple teams.
- Built micro-frontend architecture at scale with Webpack Module Federation and shared Storybook/Astro environments.
- Led Angular upgrades up to v20, dependency strategy, semantic versioning, compatibility management, and quality gates.
Earlier Engineering Roles
- Delivered production frontend platforms, micro-frontends, shared Angular libraries, design systems, and complex data-heavy applications.
- Worked on RPA/no-code automation, public service portals, fintech platforms, PWA applications, worldchess.com, and Linux-based deployments.
- Mentored engineers, drove cross-team code reviews, stabilized legacy codebases, and aligned engineering practices across teams.
Capability Matrix
Grouped by the kind of problem each capability helps solve.
Useful when a product needs reasoning, document intelligence, retrieval, structured outputs, human review, and measurable behavior.
Useful when data is sensitive, workflows are constrained, and product decisions need to respect real medical and operational context.
Useful when AI capabilities need to become usable interfaces, not hidden scripts or brittle internal demos.
Useful when prototypes need deployment paths, compatibility management, runtime adapters, test hooks, and operational ownership.
Contact
Send a role, project brief, or AI product problem. LinkedIn carries recommendations; GitHub is a public code archive.