Tarek Masryo
AI Engineer
I build production ML and Generative AI systems with measurable behavior, explicit failure modes, and reliable operation under real-world constraints.
AI Systems Engineering; Model Serving, APIs and AI Infrastructure; RAG, Retrieval and
Knowledge Systems; AI Agents, Tool Use and Execution; Evaluation, Benchmarking and System
Reliability; LLMOps, Observability and Production Operations.
View CV
About Me
I'm Tarek Masryo, an AI Engineer building production ML and Generative AI systems across model serving, retrieval, evaluation, and operation.
I work across software engineering and machine learning. I integrate models with APIs and backend services, build model-serving and retrieval infrastructure, develop RAG and agent workflows, and create the evaluation and operational tooling needed to run them reliably.
I build AI systems end to end. Model quality is only one part of the problem; data, preprocessing, retrieval, tool execution, application logic, latency, cost, and failure handling all shape production behavior. When something breaks or degrades, I trace the execution path across these layers to identify where the problem actually originates.
Evaluation and measurement are part of the engineering loop, not a final checkpoint. I use reproducible workflows, executable checks, failure analysis, and observability to compare changes and diagnose regressions, while making trade-offs across quality, latency, cost, reliability, security, and maintainability. I prefer clear interfaces, explicit failure modes, and systems that are straightforward to test, operate, and evolve.
Featured Work
Four public systems showing how I approach reliability, evaluation, and operation beyond the prototype stage.
Fraud Risk Ops Platform
Model serving · policy decisions · audit trails · operations.
Operational flow: score → policy → review → audit → monitoring
RAG QA Command Center
Retrieval diagnostics · trace inspection · configuration comparison.
Evidence: 3,824 offline QA evaluation runs · 93,375 retrieval events
LLMOps Telemetry Command Center
Reliability · latency · cost · routing · triage.
Evidence: 9,000 offline interactions · 1,595 sessions
Selected Projects
Additional work across applied ML, computer vision, analytics, and decision systems.
NLP Workspace
Advanced ML Sentiment Lab
Text classification · model comparison · ROC-AUC/PR-AUC evaluation · threshold tuning · error analysis.
Healthcare ML
Hospital Deterioration — Next 12h Early Warning
Patient-level leakage controls · Average Precision and ROC-AUC · alert-policy trade-offs.
Decision App
Short Video Intelligence Dashboard
Virality scoring · engagement KPIs · creator comparison · timing patterns · segment benchmarks.
Computer Vision App
Old Photo Restorer
User-facing image-restoration app with before/after review, presets, and batch ZIP export.
Dashboard
EV Charging Dashboard
Interactive EV infrastructure dashboard for coverage analysis and fast-DC allocation scenarios.
ML Pipeline
Pima Diabetes Pipeline
Schema validation · calibrated probabilities · threshold trade-offs.
Capabilities
Building AI systems across model integration, deployment, evaluation, and reliable operation.
Production ML Systems
Turn trained models into production services with validated inputs, calibrated outputs, clear decision logic, and reliable inference.
RAG & Retrieval Systems
Build ingestion, indexing, retrieval, reranking, grounding, and evaluation pipelines with explicit quality and retrieval checks.
AI Agents & Tool Execution
Design agent workflows with explicit execution steps, tool use, structured outputs, safeguards, and visibility into intermediate behavior.
LLM & Agent Evaluation
Build executable evaluations and benchmarks to compare behavior, surface failure modes, and catch regressions before release.
AI APIs & Model Serving
Expose ML and LLM capabilities through typed APIs and backend services with versioned artifacts, background jobs, readiness checks, and clear runtime contracts.
LLMOps & Reliability
Track quality, latency, cost, drift, and failures so changes can be compared before rollout and monitored after release.
Technical Stack
A focused set of tools for building, evaluating, serving, and operating production AI systems.
Core Stack
A focused core stack for building and operating production AI systems.
Python
FastAPI · Pydantic v2
PostgreSQL · pgvector
Redis
PyTorch
scikit-learn
Hugging Face Transformers
Sentence Transformers
LangGraph · MCP
Docker · Docker Compose
pytest · ruff
Prometheus · OpenTelemetry
GitHub Actions
Explore Engineering Toolkit
AI & Model Engineering
Training, inference, explainability, and model integration.
PyTorch
scikit-learn
XGBoost · LightGBM
Hugging Face Transformers
Sentence Transformers
SHAP
ONNX · ONNX Runtime
TensorFlow · Keras
Retrieval, RAG & Agents
Knowledge retrieval, orchestration, tool execution, and evaluation.
RAG · Hybrid Retrieval · Reranking
Embeddings · Vector Search
pgvector · FAISS
LangGraph · LlamaIndex · LangChain
Tool Use · Agent Workflows · MCP
Ragas · DeepEval
Fine-tuning
Ollama · vLLM
APIs & Application Systems
Serving AI capabilities through reliable software interfaces.
FastAPI · Pydantic v2
SQLAlchemy · Alembic
REST APIs
Redis · RQ
Background Jobs
Model Serving
Schema Validation
Typed Interfaces
Data & Retrieval Infrastructure
Storage, validation, indexing, and data workflows.
PostgreSQL · pgvector
Redis
DuckDB · SQLite
NumPy · Pandas · Polars
Pandera
pypdf · unstructured
Document Parsing
Chunking · Metadata Extraction
Reliability & AI Operations
Testing, deployment, observability, and production behavior.
Docker · Docker Compose
GitHub Actions · CI/CD
pytest · ruff · mypy
Prometheus · OpenTelemetry
MLflow
Structured Logging
Readiness · Health Checks
Caddy · Nginx
Analysis & Operator Interfaces
Analysis, debugging, review, and operational interfaces.
Jupyter · Google Colab
Plotly
Matplotlib · Seaborn
Streamlit
Gradio
Chainlit
React.js
Working With Me
How I scope, build, evaluate, and hand off production AI systems.
What kind of work do you take on?
I work on production ML and Generative AI systems, including RAG, AI agents, model serving, evaluation, AI APIs, and reliability work. I can build new systems or help move existing prototypes toward production.
How do you approach an existing system?
I first map the architecture, data flow, dependencies, constraints, and known failure points. I keep what works, simplify where possible, and make changes that can be tested and measured.
How do you decide when something is ready?
I define measurable acceptance criteria around the things that matter for the system: quality, retrieval performance, failure rates, latency, cost, reliability, or review load. A good model metric alone is not enough.
What does delivery include?
Usually code, tests, configuration, reproducible setup, evaluation artifacts, documentation, and deployment guidance. Monitoring and review workflows are included when the system needs them.
How do you handle reliability?
I use explicit contracts, validation, regression tests, versioned artifacts, structured logging, health checks, and clear failure behavior so problems are easier to reproduce and diagnose.
What happens after delivery?
After delivery, the system should be clear to use, easy to maintain, and straightforward to operate. I document the important assumptions and known limits, and I can stay involved with monitoring, stabilization, or further improvements when needed.
