Senior AI/ML engineer with deep expertise in custom model training, agentic architectures, and real-world deployment — helping businesses harness AI from first strategy to production.
I'm Casey Turtel, a senior software engineer specializing in AI and machine learning. I hold a B.S. in Electrical & Computer Engineering from CU Boulder with concentrations in Applied Mathematics and Philosophy.
At CableLabs, I architected Network Comprehension — a custom encoder-only transformer with LSTM positional encoding for real-time network traffic analysis on resource-constrained hardware. I've built multimodal agentic systems, fine-tuned LLMs for client-specific personalities, and shipped everything from HSM firmware to high-speed FPGA order books.
I've also developed a growing library of specialized Anthropic Claude skills — domain-specific AI capabilities deployed as production tools for architecture, law, medicine, engineering, and more.
Whether you're just exploring what AI can do for your business or ready to build custom models, I bridge the gap between AI hype and real, measurable outcomes.
Not sure where to start? I assess your business operations and identify the highest-leverage opportunities for AI — with honest tradeoffs, ROI estimates, and a prioritized roadmap you can actually execute.
When off-the-shelf models aren't enough, I design and train specialized models tailored to your data and domain. From encoder-only transformers to fine-tuned LLMs, I build models that understand your world.
I build autonomous AI agents and retrieval-augmented generation pipelines that reason, plan, and act — giving your LLMs accurate knowledge and the ability to use real tools.
I specialize in building and deploying Claude Skills — domain-specific capabilities that extend Claude's intelligence into specialized professional fields, ready to deploy as APIs or embedded tools.
Most industries are still in the early innings of AI adoption. I help you understand the realistic landscape — what your competitors are doing, what's actually working, and where the asymmetric opportunities are right now.
A model that can't ship is worthless. I design inference pipelines, containerized deployments, and monitoring systems that keep your AI running reliably — from cloud to edge.
Contracts, reports, emails, manuals — AI can read, summarize, classify, and extract structured data at scale.
Replace repetitive knowledge-worker tasks with agents that reason, decide, and act — freeing your team for higher-value work.
AI-powered support, lead qualification, and personalization that scales without scaling headcount.
Make your company's institutional knowledge instantly queryable. RAG systems that actually understand your docs.
Custom ML models trained on your data to forecast demand, detect anomalies, and surface actionable signals.
Real-time pattern recognition on network traffic, logs, and user behavior to catch threats before they escalate.
LLMs grounded in your industry's terminology, regulations, and best practices — not generic chatbots.
AI-augmented development workflows, code review, documentation generation, and test coverage at scale.
Claude Skills are domain-specific AI capabilities built on Anthropic's Claude — combining deep contextual knowledge, expert-level prompting, and specialized tooling to give Claude genuine expertise in professional fields. I've developed a library of 17+ production skills spanning engineering, law, medicine, mathematics, and spatial reasoning. Each skill can be integrated into business workflows, internal tools, or client-facing applications.
A track record of shipping real AI systems across telecommunications, fintech, and emerging tech.
Side projects that pushed into new territory — from FPGA arbitrage engines to offline AI survival kits.
Contracted by a well-known TV personality to fine-tune Llama 3.3 on their public and private content (speeches, articles, TV appearances). Used GCP for training pipelines and deployment. Now used by their assistant to respond to emails and generate blog posts.
A completely offline, speech-capable AI assistant with RAG over a massive local knowledge base — Wikipedia, survival guides, medical textbooks, carpentry references. Runs Llama 3.1:8b on a Jetson Nano, TinyLlama on Raspberry Pi. Spawned a "traffic lawyer in a box" variant for use during traffic stops.
Custom ML model combining stock market data with semantic analysis of news, blogs, and tweets to predict price movements. Used reinforcement learning, trained on 5 auto-manufacturing stocks, tested on a 6th. Achieved 15% returns over 2 months in testing.
Arbitrage order book implemented in Verilog using a custom networking protocol. Deserializes orders containing market, asset, buy/sell flag, and price into local memory. Identifies cross-exchange spread opportunities and executes fills in just two clock cycles after message receipt.
Built for a bar: a device that detects TV commercials in real-time and replaces them with the bar's own ads. Fine-tuned ResNet50 on manually collected data with a hybrid similarity-search + model approach for robust detection.
Capstone team lead and sole programmer. Bike attachment that automatically shifts gears based on pedal cadence and torque. Implemented in C on ATMega, reading hall effect and torque sensors, controlling a linear actuator to tension the drivetrain.
From basketball leagues to fashion platforms — building things beyond the day job.
A platform for embedding third-party social apps — Letterboxd, Spotify, Strava, Goodreads — directly into a shareable profile page. Friends can stay up to date with what you're reading, watching, and exercising without jumping between apps.
For influencers, it creates a single destination to advertise, connect with fans, and showcase activity across every platform they're active on — all in one scroll.
A relegation-based basketball league that brings professional-style competitive structure to recreational play. Compete, earn your way up — or fight to stay in. The table never lies.
A platform connecting stylists — and anyone with great fashion sense and free time — with people in need of personalized fashion advice. Your style, on demand, from real people who love it.
Whether you need a strategic AI roadmap, a custom trained model, or a production-ready deployment — I'm happy to talk through what's actually possible for your situation.
Reach out about any of the following — or something else entirely.