AI Engineer
Kevin King
I build and deploy production AI systems for enterprise clients: creative workflow automation, agentic orchestration, RAG, and full-stack apps, from requirements to production.
Selected Work
Production AI and agentic systems
Law Firm Staff Competency Platform
Forward-deployed engagement: embedded with a personal-injury law firm to build their staff training and assessment platform end to end. Study materials with TTS, a 70-question timed exam, LLM rubric grading, admin gap analysis, and branded PDF reports — in production with active staff.
Study module serves firm doctrine with ElevenLabs TTS. Exam engine handles 70 questions across five formats with autosave. Objective questions scored server-side (answer key never in client bundle); written answers graded by LLM against model answers via strict-JSON rubric prompts, results cached for stable re-reads. Admin dashboard surfaces per-topic gaps and AI team insights; a data-to-PDF pipeline renders branded readiness reports for leadership. GitHub Actions typecheck gate catches regressions before they reach the customer.
creative-automation-pipeline
End-to-end generative AI pipeline for creative teams: campaign brief intake, brand-voice modeling, and automated production of newsletter content, ad copy, and video ad variants at scale.
Brief intake layer parses campaign inputs into structured schemas. Brand-voice embeddings constrain generation. Parallel LLM calls produce channel-specific variants (newsletter, static ad, video script). Validation loops score brand consistency and platform format compliance before output.
Ecliptic Intelligence and Sterling
Multi-tenant Chief of Staff dashboard for founders and law firms. Sterling reduced daily founder operations triage from 60 minutes to 5. Full funnel on Stripe, Resend, Meta CAPI, and Clerk auth.
Event-driven multi-agent backend: specialized Python sentinels (Fire, Inventory, CFO, Operations) each own a state machine with retry logic and self-critique loops. FastAPI gateway, PostgreSQL (Neon) for persistence, LangChain and Mastra for orchestration, Vercel edge deployment.
Aurum
Agentic customer-acquisition infrastructure with enrichment, fit-scoring, and personalized multi-touch outreach, human-approved and fully traced.
LangGraph pipeline: enrichment agent pulls and normalizes prospect data, fit-scoring agent evaluates against ICP criteria, personalization agent generates sequenced outreach. Human-in-the-loop approval gate before any send. Full trace logging across every node for auditability.
ML, data, and systems
HIPAA-compliant legal RAG system
Retrieval system for personal injury case files deployed inside a client AWS environment. Reduced case review cycles by 70 percent, improved retrieval accuracy by 30 percent, and cut demand-letter prep effort by 60 percent.
Airflow ingestion pipeline handles OCR, chunking, and embedding of case documents. Hybrid vector + keyword retrieval (Pinecone + Kendra + Aurora). Claude via Bedrock for summarization and demand-letter drafts. IAM-scoped access, PHI controls, all compute inside the client VPC.
muse-churn-prediction
Customer churn prediction and LTV forecasting on real Shopify data. Improved demand forecasting accuracy by 22 percent and reduced stockout events by 18 percent.
ETL pipelines in Cloud Composer and Airflow pull Shopify and analytics data into BigQuery. TensorFlow and Scikit-Learn for churn classification and customer segmentation. Prophet and ARIMA time-series models for demand forecasting. BigQuery ML for cohort LTV analysis.
bayguard
AI-powered real-time Chesapeake Bay health monitoring system. Fuses live federal data streams with a semantic institutional memory layer that retains and reasons over historical environmental patterns.
Federal data ingestion (NOAA, EPA, USGS) feeds a signal analysis layer for anomaly detection. A RAG system built over embedded historical monitoring reports, regulatory filings, and scientific literature gives the system institutional memory. LLM query interface for natural-language environmental Q&A.
easement-sentinel
Cloud-hosted web app and CLI that turn field photos and parcel data into litigation-grade conservation-easement monitoring evidence packets.
Vision model pipeline processes field photographs for change detection against baseline imagery. Parcel data and GIS coordinates are cross-referenced against easement terms. Output layer assembles timestamped, annotated evidence packets formatted for legal submission.
Supply chain optimization
Mixed-integer linear programming model for multi-echelon supply chain optimization, cutting total cost across sourcing, inventory, and routing constraints.
MILP formulation in Python with Gurobi solver. Models supplier selection, inventory positioning, and distribution routing as a unified optimization problem. Constraint sets cover capacity, lead times, demand variability, and cost objectives across multiple echelons.
Stack
LLMs
Orchestration
Languages
Vector / Retrieval
Infrastructure
Marketing
Creative
Experience
Founder, AI Solutions Architect
Ecliptic Intelligence
Python, FastAPI, Claude, GPT-4/5, Gemini, TypeScript, PostgreSQL, Vercel, LangChain, Mastra
- Built Sterling, an AI Chief of Staff deployed in a live DTC brand — reduced daily ops triage from 60 minutes to 5 through intelligent briefs, priority surfacing, and automated monitoring.
- Deployed event-driven Python architecture with specialized sentinels managing state, retry logic, and self-critique loops. Caught an overdue supplier payment and flagged inventory risk 2 days before a potential stockout within the first week.
- Shipped newsletter, ad copy, and video ad generation agents for active marketing channels including a newsletter at 1,100+ subscribers.
AI Systems Engineer
Regional Legal Services Firm
Python, AWS Bedrock (Claude), Pinecone, Airflow, PostgreSQL Aurora
- Forward-deployed inside a client AWS environment: led requirements solutioning with firm partners and senior attorneys, blueprinted the architecture, and owned deployment end-to-end with zero access to external infrastructure.
- Reduced case review cycles by 70 percent, improved retrieval accuracy by 30 percent using hybrid vector and keyword search (Pinecone, Aurora, Kendra), and cut demand-letter preparation effort by 60 percent.
- Automated ingestion, OCR, chunking, and embedding workflows using Airflow and Cloud Composer. Implemented IAM-restricted Bedrock access patterns with encryption and PHI exposure controls for HIPAA compliance.
Marketing Operator, Data Scientist
Muse of the Moon
Python, TensorFlow, BigQuery, Airflow, Shopify, Meta Ads Manager, Klaviyo
- Grew a DTC jewelry brand from early-stage to multi-six-figure revenue: 50,000+ followers, 5,000-customer email list, thousands of organic sales across Etsy and Shopify.
- Improved demand forecasting accuracy by 22 percent and reduced stockout events by 18 percent using TensorFlow, Prophet, and ARIMA models on BigQuery.
- Built ETL and ELT pipelines in Cloud Composer and Airflow, eliminating 10 hours per week of manual reporting. Ran Meta ads end-to-end against ROAS targets.
About
I am Kevin King, an AI engineer based in coastal Virginia. I build production AI systems for founder-led businesses and law firms, with a foundation in machine learning and optimization.
Outside of AI engineering, I run two creative brands: Lightbender Visuals (light painting and music industry creative work) and Wild Eye Visuals (Chesapeake Bay wildlife photography). Production use of the full Adobe Creative Suite across both, including Photoshop, Lightroom, After Effects, and Premiere. I understand creative workflows from both sides.
M.S. Business Analytics, William and Mary. B.S. Business Administration, Old Dominion University. TensorFlow Developer Certified. Google Data Analytics Professional Certified. AWS Solutions Architect, in progress.
Get in touch
Available for new engagements. If you're building something that needs production AI, reach out.