const kaveh = { role: "Lead Staff Software Engineer | Data Platforms, ML Systems & GenAI", experience: "20+ years building software; 12 years at Comscore", trajectory: "data platforms → ML systems & GenAI", education: [ "M.Eng Computer Science (ML) @ Virginia Tech", "M.Sc Software Engineering @ George Mason University" ], currentFocus: [ "production data platforms", "AI-assisted pipeline migrations", "PyTorch ML/DL experimentation", "RAG evaluation and GenAI infrastructure" ], operatingMode: "prove parity, automate delivery, keep the architecture legible" };
▸Building production data platforms, ML systems, and GenAI applications.
A self-hosted, source-configurable, multi-disciplinary engineering platform — GenAI, ML, and data. Bundles 30+ open-source services (Ollama, LiteLLM, Supabase, n8n, Airflow, Spark, Weaviate, Neo4j, JupyterHub, Ray, ComfyUI) behind a Kong gateway, with an interactive TUI wizard, selectable tracks, and one-command launch.
Seven RAG strategies — vanilla, hybrid, contextual, LightRAG graph, agentic, n8n-adaptive, and an experimental lazy-graph — served side by side as OpenAI-compatible endpoints on a fully-local atlas stack, with reproducible Ragas + LLM judge-panel evaluation that measures which approach wins on which kind of question.
A lightweight PyTorch toolkit for training, evaluating, and visualizing modern neural networks, with first-class graph neural network support (GCN/GraphSAGE/GAT). Covers feed-forward, graph, decoder-only LM, diffusion, JEPA, and mixture-of-experts models; parameter-efficient fine-tuning (LoRA/DoRA/IA3), quantization, and pruning; and ONNX/GGUF/Ollama/Hugging Face export.
A hands-on PyTorch notebook portfolio where each notebook runs end to end and is CI-validated via NNx — spanning classical ML, GNNs, NLP, transformer LMs with BPE, DDPM diffusion, DPO, JEPA, MoE, PEFT, quantization, pruning, and knowledge distillation.
An Apache Iceberg lakehouse data-engineering lab built on atlas — 19 Spark scenarios in Scala/PySpark across a bronze/silver/gold medallion, two CI-built Maven Spark apps run by Airflow, Trino SQL/BI, and Redpanda streaming.
A hierarchical, lifecycle-aware MVVM framework. One language-neutral spec drives five idiomatic, full-parity implementations (C#, Python, TypeScript, Swift, and Rust) cross-checked by 401 shared conformance scenarios; reactive-first (Rx/RxJS/ReactiveX/Combine) and published to NuGet, PyPI, npm, SwiftPM, and crates.io.
A cross-platform TUI for select AWS and S3-compatible services (macOS, Linux, Windows), powered by Textual + VMx MVVM.
A fuzzy multi-criteria (TOPSIS) decision-support tool for software architects with three parallel implementations (TS+Tauri+Svelte, C#+Avalonia, Python+NiceGUI) kept in lockstep by a shared spec; built on VMx.
A GenAI language-learning app: AI-persona chat, translation puzzles scored via embeddings, streaming content generation, and skill assessments; FastAPI + Postgres + Ollama backend, strict-MVVM NiceGUI + VMx frontend.