Languages: Python, Java, C++, JavaScript, TypeScript, SQL, JSON
ML/NLP/GenAI: GPT-4o/4/3.5, LangChain, Transformers, spaCy, Whisper, VAEs, CNNs, RNNs, LSTMs, distilBERT, GANs, Keras, SparkML
Frontend: React, TypeScript, HTML/CSS, Flask, Django, Streamlit
Infrastructure: Docker, Kubernetes, CI/CD, AWS, GCP, Azure, Git, Agile, DataOps
Analytics & BI: DuckDB, Altair, Power BI, Tableau, DataRobot, H2O, Microsoft Fabric, Databricks, Jupyter Notebook
Data & Pipelines: SQL Server, Oracle, NoSQL, BigQuery, Redshift, FAISS, SSIS, ActiveBatch
Other: FastAPI, RESTful APIs, MPC

I build full-stack AI products that bring models into real-world use—merging UX, analytics, and privacy-first design. I prioritize stacks that are tightly engineered, cost-effective, and highly transparent. I like clean architecture and user-mesmerizing UIs (or at least clean, user-friendly, non-offensive frontends when strapped for time). I always code with the next guy in mind—even if that next guy is future me.

My open-source projects include GenAI-native BI platforms, LLM-driven call center analytics, and anomaly detection tools for automotive diagnostics. These systems combine LangChain, GPT-4 (and GPT-3.5 when affordable token cost is a priority), semantic search, vector databases, custom pipelines, and real-time analytics layers using DuckDB and Altair. I’ve shipped local-first apps with PII redaction (spaCy + Presidio), live transcription (Whisper), and smart dashboards (React + TypeScript) that surface context-aware insights. I connect LLMs to structured and unstructured data sources, and build end-to-end workflows that actually work under real-world constraints.

I design LLM-powered assistants for real users—not demos, but multi-capable agents built to carry real workloads. For example, I built a full-stack AI assistant for under-staffed nonprofits, integrating eight distinct voices and capabilities into a single chat interface. It could engage in natural language dialogue, search for relevant research, clean and rank sources, and auto-insert citations into legal documents or persuasive outreach it generated. All prompt-engineered, tightly scoped, and built to serve mission-driven teams with limited staff and no technical background.

I build public proxies for private systems—stripped of proprietary data but still engineered with real-world logic, constraints, and value. They’re field-adjacent, practically useful tools that stand in for things most people can’t normally access. Where I can’t release proprietary code or OEM data, I build accessible tools that mirror the logic and impact of those systems. I’ve open-sourced a trained YOLOv8 model for auto body damage detection—tuned to work with small datasets and runnable on Google Colab. An enhanced version of that model now runs in BMW service environments to support pre-loaner vehicle inspections. I’ve also released a BMW MHD log anomaly detector—a field-useful diagnostic tool for tuners and enthusiasts. It flags issues in real driving data using Isolation Forests and binary spike indicators (AFR, throttle, timing, etc.), giving public users a slice of what’s normally locked inside internal engineering systems.

I also use advanced analytics to extract insights from large, complex datasets and drive decision-making in business, research, and automotive (OEM) environments. I work with cross-functional teams to solve real problems, build predictive models, and deploy scalable AI that boosts efficiency, improves user experience, and fuels growth. I create AI strategies tied to business goals and find opportunities to apply machine learning for impact.

I collect, clean, and prep large datasets, apply statistical methods and ML algorithms, and build predictive models. I tune and evaluate models, push them to production, and integrate them into workflows for scale and reliability. I manage AI risk—ethics, privacy, compliance, and security—and build mitigation plans.

In education and research, I build generative AI apps and recommender systems for automated research, personalized learning, and semantic search. I use LangChain, Hugging Face, VAEs, autoencoders, and diffusion models. I connect these systems to structured databases and cloud platforms (AWS, Azure, Google Cloud), and build scalable research platforms using Docker, Kubernetes, and CI/CD. I also build custom tools for research institutes—like using computer vision models to read cuneiform and assist ANE researchers in deciphering ancient scripts such as Elamite.

In automotive, I build AI/ML models using deep learning architectures (RNN, CNN, LSTM) and computer vision with OEM datasets. I apply clustering, gradient boosting, anomaly detection, and neural networks. I deploy full AI pipelines using TensorFlow, PyTorch, and no-code platforms like DataRobot and H2O.

I code in Python, Java, C++, JavaScript, and TypeScript. I work with SQL, Oracle, SQL Server, and NoSQL databases, and routinely handle structured data in JSON. I develop UIs and full-stack apps using HTML, CSS, and modern frontend frameworks. My work spans classical ML, NLP, and LLM-based systems—from spaCy pipelines to GPT-powered tools. I track KPIs to measure impact, fine-tune strategies, and lead AI adoption. I build dashboards using both Power BI and custom frontends (React + TypeScript), and integrate them with backend analytics, cognitive services, and Microsoft tools like Azure Speech for richer insight delivery. I also advise on AI product strategy, focusing on real-world applications in business.