Machine Learning Systems
Churn, inventory demand, and risk models designed with clear objectives, calibrated outputs, and deployment-ready evaluation artifacts.
AI Systems Portfolio
Designing and shipping production-grade ML, NLP, and RAG systems.
This portfolio highlights end-to-end AI work: supervised ML pipelines, document intelligence with retrieval-augmented generation, and observability designed for real operational systems rather than toy demos.
MLChurn, forecasting, risk modeling
NLPClassification, QA, summarization
RAGPDF ingestion, retrieval, grounded answers
Systems at a glance
Production-oriented AI work organized around reusable services, consistent interfaces, and observable pipelines.
ML SystemsNLP ToolingRAG WorkflowsObservability
Featured engineering themes
Modeling, retrieval, structured UX, and source-grounded response generation brought together in a single portfolio.
What this site is built to show
That the work goes beyond notebooks: clear architecture, repeatable workflows, and systems that can actually be operated.
Systems overview
AI systems built end to end
A concise view of the core domains covered in this portfolio: supervised ML, NLP tooling, retrieval-augmented generation, and observability for production pipelines.
NLP Systems
Classification, NER, summarization, translation, and QA workflows exposed through consistent service patterns and reusable interfaces.
RAG Workflows
PDF ingestion, chunking, embeddings, retrieval, reranking, and grounded answers assembled into an auditable document assistant.
Observability
Metrics, confusion matrices, drift signals, and monitoring artifacts used to keep systems interpretable after deployment.
Featured systems
Representative AI systems
A small sample of end-to-end systems that combine modeling, retrieval, and product-grade presentation.
Applied MLRegression
AI Building Estimator
Structured ML workflow that estimates building costs using tabular features, calibrated risk bands, and evaluation-ready artifacts.
Risk bandsEvaluation
Explore system →RAGDocument intelligence
RAG Document Assistant
Document-grounded QA over PDFs with retrieval, reranking, chunk attribution, and source-aware answer generation.
EmbeddingsRetrievalGrounded answers
Open assistant →NLPLLM workflows
NLP Systems Showcase
A composable workspace for experimenting with classification, NER, summarization, translation, and QA over realistic text inputs.
Text classificationNERQA
View showcase →Engineering approach
How systems are structured
The portfolio is organized around complete AI systems rather than isolated models, with clear separation between data, retrieval, reasoning, response, and monitoring layers.
- Projects are framed as end-to-end systems with explicit inputs, outputs, and operating assumptions.
- Interfaces are kept stable so individual components can evolve without collapsing the entire flow.
- Monitoring and evaluation are treated as first-class design concerns, not afterthoughts.
Data Layer
Structured and unstructured inputs such as tables, logs, and PDFs prepared into model-ready representations with clear contracts.
Retrieval Layer
Vector stores, metadata filters, and ranking strategies tuned for relevance, latency, and traceability across search and RAG workflows.
Reasoning Layer
Classical models and LLM-based components wired through reusable services that encapsulate prompts, business rules, and safety checks.
Response Layer
Stable response schemas, robust error handling, and UI patterns designed for dependable downstream consumption.
Monitoring Layer
Metrics, drift indicators, evaluation artifacts, and operational dashboards that make regressions visible instead of silent.
RAG demo
Document-grounded question answering
Upload a PDF, let the system ingest and index it, then ask questions that are answered strictly from the document with visible sources and retrieval-backed context.
PDF ingestion with text extraction and chunking optimized for retrieval.
Embeddings, vector search, and optional reranking to select relevant passages.
Answer generation that keeps citations and source visibility front-and-center.
Frontend workflow designed to feel like a usable product rather than a raw technical demo.
Tech stack
Technologies used across the portfolio
The focus is on pragmatic, production-ready tools: a small, well-understood stack that supports experimentation, stable deployment, and operational clarity.
PythonFastAPIPydanticUvicornREST APIscikit-learnpandasNumPyjoblibClassificationRegressionTime seriesRisk modelingOpenAI APIOpenAI EmbeddingstiktokenText classificationNERSummarizationTranslationQuestion AnsweringFAISSPyPDFVector searchChunkingRerankingNext.jsReactTypeScriptTailwind CSSAxiosDockerpython-dotenvMetrics & observability
Selected projects
Project depth across ML, NLP, and document intelligence
A subset of projects highlighting supervised learning, document-grounded retrieval, and NLP tooling built with end-to-end thinking.
Applied ML
AI Building Estimator
Cost estimation system for building projects with explainable risk bands and evaluation artifacts.
Read case study →ML system
Customer Churn Prediction
End-to-end churn pipeline with engineered features, probabilistic outputs, and observability-ready metrics.
See details →Forecasting
Inventory Demand Forecasting
Multi-horizon demand forecasts designed to support inventory planning and scenario analysis.
Explore project →Risk modeling
Pension Risk Modeling
Structured risk estimates for pension portfolios with ROC/PR curves, confusion matrices, and calibrated evaluation views.
View project →NLP
NLP Pipeline Showcase
Composable workflows for classification, summarization, translation, and question answering.
Open showcase →RAG
RAG Document Assistant
Document-grounded QA over PDFs with retrieval, reranking, citation visibility, and answer traceability.
Try demo →About
Mohammad Amiri — AI & ML engineer
Focused on building end-to-end AI systems that combine robust modeling, retrieval workflows, and operational clarity. The emphasis across this portfolio is on systems that can be deployed, measured, and improved over time.
Production-grade AIApplied MLNLP SystemsRAG WorkflowsSystem Architecture
Continue into the systems
Explore the portfolio like a product, not a document.
Review the ML systems, open the RAG document assistant, browse selected projects, or continue to the resume and contact paths. Each page is designed to show how the components fit together.