Junior AI Engineer in Milan building agentic AI systems, multi-agent workflows, LLM-powered document intelligence, backend APIs, and production-oriented machine-learning applications. I focus on transforming complex manual processes into modular, testable, and measurable AI products.
Designing specialized AI agents with clear responsibilities, tools, structured outputs, validation rules, and auditable execution flows.
Coordinating multiple LLM agents for extraction, classification, validation, cross-checking, and decision-support workflows.
Transforming unstructured CVs, legal files, financial records, and technical documents into reliable, schema-driven data.
Using embeddings, vector stores, and contextual retrieval to improve model grounding and access to domain-specific knowledge.
Building FastAPI services, REST endpoints, asynchronous workers, database integrations, automated tests, and containerized environments.
Integrating predictive models, feature pipelines, asynchronous inference, SHAP explanations, and interactive data applications.
Developed Phase 1 of a multilingual CV-processing platform. Designed specialized agents for personal information, education, skills, and work experience; implemented multi-agent orchestration, prompt engineering, tool calling, structured JSON outputs, validation, fallback logic, API testing, code refactoring, and a Streamlit review interface. The workflow reduced candidate-processing time from one–two weeks to minutes.
Professional project · private repositoryContributed to an AI due-diligence solution developed for client Nivara. Built and refined multi-agent workflows for document classification, structured extraction, validation, discrepancy detection, and cross-document consistency checks across financial, legal, and real-estate records. Applied schema-driven outputs, API integration, automated testing, and human review for uncertain results.
Client project for Nivara · private repositoryContributed to a multi-service AI/ML platform supporting queued vehicle-price forecasting, background task execution, job-state tracking, result persistence, SHAP-based explanations, Streamlit dashboards, and LLM-assisted valuation. Worked across API flows, worker integration, testing, and production-style architecture.
Professional project · private repositoryAnalyzed IoT and operational data, developed preprocessing and feature-engineering pipelines, supported anomaly-detection models, and created KPI visualizations. Also implemented a transformer-based multilingual translation workflow for technical and operational documentation, reducing manual translation effort and turnaround time.
Master’s thesis internship projectSelected professional projects were developed in company or client environments. Their source code is private, but architecture, responsibilities, technologies, and measurable outcomes are summarized here.
Achieved approximately 98% extraction and classification accuracy while processing around 50 diverse CVs in AIRING Phase 1.
Created eight agents for AIRING and worked with more than ten specialized agents in the MyCredit AI workflow.
Automated structured candidate processing that previously required one to two weeks of manual work.
Helped compress an approximately four-week financial and document-analysis workflow to around fifteen minutes.
Contributed to API endpoints and backend workflows supporting extraction, validation, job execution, and user review.
Worked across AIRING, MyCredit AI for Nivara, and LIVREA, covering document intelligence, agentic workflows, and ML inference.
I am open to Junior AI Engineer, Generative AI Engineer, LLM Application Engineer, and Machine Learning Engineer opportunities in Italy or international teams.
mostafa.hashemiyan@mail.polimi.it