const kaveh = { role: "Lead Staff Software Engineer | Data, ML & GenAI Engineering", experience: "15+ years building software; 12+ at Comscore", trajectory: "large-scale data engineering → ML systems & agentic AI workflows", education: [ "M.Eng Computer Science (ML) @ Virginia Tech", "M.Sc Software Engineering @ George Mason University" ], currentFocus: [ "production Scala/Spark pipelines", "AI-assisted migration workflows", "RAG/LLM systems and evaluation", "open-source ML/data platforms" ], operatingMode: "prove parity, automate delivery, keep the architecture legible" };
▸Building production data, ML, and GenAI systems that validate and automate legacy workflows.
Self-hosted, source-configurable, multi-disciplinary engineering platform — Gen-AI, ML & data — composable from a single Docker Compose stack. 30+ OSS services (Ollama · LiteLLM · Supabase · n8n · Airflow · Spark · Weaviate · Neo4j · JupyterHub · Ray …) pre-wired behind Kong. Interactive TUI wizard, selectable tracks, one-command launch.
Six RAG strategies — vanilla, hybrid, contextual, graph, agentic, and n8n-adaptive — served as OpenAI-compatible endpoints selectable from the Open WebUI model picker on a fully-local Atlas stack, with a reproducible LLM judge-panel harness that measures which approach wins on which kind of question.
Lightweight PyTorch toolkit for training, fine-tuning, and exporting modern neural nets. FFN + GNN + decoder-only LM + diffusion + JEPA + MoE; PEFT (LoRA/DoRA/IA3/Prefix/Prompt), PTQ/QAT, pruning, surgery; ONNX/GGUF/Ollama/HuggingFace Hub interop. Dataclass-configured runs with fluent builders, automatic checkpointing, Plotly viz.
Self-contained notebook portfolio for PyTorch: classical ML, GNNs, NLP, transformer LM + BPE, DDPM diffusion, DPO, JEPA, MoE, PEFT (LoRA/DoRA), PTQ/QAT quantization, pruning, KD. Each [task]-[dataset]-[model]-[framework] folder runs end-to-end. CI-validated via shared nnx toolkit.
Full-featured Apache Iceberg lakehouse data engineering lab with 19 Spark scenarios in Scala and PySpark, 2 CI-verified Maven Scala Spark applications built by Jenkins and orchestrated by Airflow, Trino SQL and BI, Redpanda streaming, and a medallion architecture across bronze, silver, and gold layers using Docker Compose to run the Atlas platform.
Hierarchical, lifecycle-aware MVVM framework. One language-neutral spec, four idiomatic flavors — C#, Python, TypeScript (full parity) + Swift (documented subset) — cross-checked by 232 shared conformance scenarios. Reactive-first (Rx · RxJS · reactivex · Combine). Published on NuGet, PyPI, npm, SwiftPM.
Cross-platform Norton-Commander-style TUI for AWS and S3-compatible services. Powered by Textual + VMx MVVM. Runs on macOS, Linux, and Windows.
Fuzzy multi-criteria decision analysis (TOPSIS) for software architecture. Three v1.0 implementations — TypeScript+Tauri+Svelte, C#+Avalonia, Python+NiceGUI — kept in lockstep by a shared language-neutral spec and a 1e-9-tolerance cross-impl conformance corpus. Built on VMx MVVM. Second-gen rebuild of the original Silverlight GuideArch.
GenAI language-learning app — chat with AI personas, polyglot translation puzzles scored via embeddings, streaming content generation/rewrite/review, skill assessments. FastAPI + Postgres + Ollama backend, strict-MVVM NiceGUI + VMx frontend, full Docker Compose stack.