Jay HunterI architect and ship real systems — solo, with agentic development.
Software engineer specializing in AI and machine-learning platforms. I use agentic AI development to design, build, and operate production systems end-to-end — from clean-architecture backends to ML pipelines and full-stack products.
- 6
- production platforms
- .NET 10
- primary stack
- Solo
- architect & builder
Selected work
Production systems, built end-to-end
Each of these is a real, working platform I designed, built, and operate — spanning AI, machine learning, automation, and full-stack product work.
InsightAide
AI-driven investment research & analysis platform
AI Platform · Machine Learning
A solo-built platform for quantitative investment research and analysis on .NET 10, structured as a 25-project clean architecture (Foundation → Services → Engines → Presentation). It runs an in-process ML.NET pipeline for signal generation and ranking, with a dual-LLM router that dispatches work across the Claude and OpenAI APIs for automated hypothesis generation. Research and analysis only — no live trading.
- Dual-LLM router that dispatches claude-* and gpt-* models across Anthropic and OpenAI, logging per-call token usage and cost to PostgreSQL
- 117-feature ML.NET pipeline with purged K-fold walk-forward validation and time-based embargoes to prevent lookahead bias
- LambdaRank cross-sectional ranking model, gated on an out-of-sample spread t-stat before a model is saved
- 25-project clean architecture on .NET 10, deployed on Railway via Docker with PostgreSQL + pgvector storage
Hunter Metric
Applied ML platform for predictive sports analytics
Machine Learning · Predictive Modeling
An applied-ML platform that models game outcomes across six professional sports. Each sport has a native win-probability model feeding a shared, Platt-calibrated probability assembler, backed by a multi-family ensemble (logistic regression + LightGBM + ELO) with agreement scoring. Models are evaluated with walk-forward backtesting and promoted through a shadow → active governance pipeline, so none reaches production without proving calibrated skill against real outcomes.
- Six native per-sport win-probability models feeding a shared Platt-calibrated probability assembler
- Multi-family ensemble (logistic regression + LightGBM + ELO) with correlation-aware agreement scoring and outlier detection
- Walk-forward backtesting with expanding and rolling folds and strict chronological-leakage invariants
- Shadow → active model governance: models accrue data silently and only graduate after calibration and walk-forward gates pass
BookingAide
Full-stack booking & scheduling platform
Full-Stack SaaS
A full-stack booking and scheduling SaaS — an ASP.NET Core 8 API behind a Next.js 16 / React front end. It handles appointments, rentals, and events with Stripe payments, automated email (Resend) and SMS (Twilio) reminders, and multi-staff, multi-location availability, with double-booking prevented at the database level.
- Database-level double-booking prevention via a PostgreSQL EXCLUDE constraint over time ranges — no app-level locking
- Idempotent Stripe payment handling keyed on payment-intent IDs to survive webhook redelivery without double-charging
- Tiered availability engine: staff hours and blackout dates override business defaults, with location-aware conflict detection
- Automated email + SMS reminder pipeline driven by a background worker with per-send idempotency markers
How I build
Agentic development, done like an engineer
Agentic development means using AI agents as a force multiplier across the whole lifecycle — not as an autocomplete. The hard part isn't generating code; it's designing the system, orchestrating the agents, and verifying the output. That's where I focus, and it's how one person ships work that normally takes a team.
Architect first
I lead with system design — clean/hexagonal architecture, clear domain boundaries, and testable seams. Agents write more code faster, so the design discipline is what keeps it maintainable.
Orchestrate, don't just prompt
I direct fleets of AI agents across planning, implementation, review, and verification — decomposing work, running parallel tracks, and adversarially checking output before it lands.
Ship and operate
These aren't demos. They're deployed platforms with real data, ML pipelines, publishing loops, and governance — built, monitored, and iterated by one person.
Verify relentlessly
Walk-forward evaluation, shadow deployments, calibration checks, and code review gates. Speed only matters if the result is correct — so correctness is engineered in, not hoped for.
Stack & integrations
The tools I ship and operate with
End-to-end delivery: I don't just write the code — I wire up the deployment pipelines, payment and email integrations, and data layer that make a system production-ready.