Art Turner
AI Application Engineer · GenAI / RAG · Python · AWS
Data-driven by default. Relentless by nature.
- • Design and deploy RAG pipelines, LLM orchestration, and multi-agent workflows end-to-end.
- • Strong Python foundation with a production mindset: containers, APIs, evaluation, CI/CD.
- • AWS-certified across 7 credentials including GenAI Developer (Professional) and ML Specialty.
Projects
Selected work (problem → approach → result).
Education-Domain LoRA Fine-Tuning Pipeline
End-to-end fine-tuning, evaluation, and deployment pipeline for a K-12 science Q&A model, orchestrated with Dagster and deployed to a live SageMaker endpoint.
- Fine-tuned Llama 3.2 1B Instruct with LoRA on 7,666 K-12 science questions (OpenBookQA + SCIQ), achieving a +22 percentage point improvement in Exact Match accuracy (54% to 76%) over the base model
- Built a dual evaluation framework: Exact Match scoring on 100 held-out questions plus LLM-as-Judge (Claude) scoring Correctness, Pedagogy, and Conciseness on a stratified 20-sample subset. Fine-tuned model scored 3.27/5 overall vs. 2.20/5 base
- Orchestrated the full pipeline with Dagster (data prep, SageMaker training, SageMaker evaluation, deployment) with a quality gate that blocks deployment if the fine-tuned model does not outperform baseline
- Deployed the merged LoRA adapter as a live SageMaker real-time endpoint (ml.g5.xlarge), verified with 3/3 correct live inference predictions
Intelligent Retrieval & Advisor-Assist System
RAG-powered decision-support application with conditional routing, grounding verification, and provider-agnostic LLM integration.
- Multi-node RAG pipeline in LangGraph with conditional routing across five query types and configurable retrieval strategies
- Hallucination mitigation via token-overlap grounding verification, retry/refuse loop, and dual confidence thresholds
- Evaluation framework with groundedness, relevance, and refusal-correctness scorers
- Page-level citation system with custom PDF extraction pipeline preserving chapter and page metadata through chunking and retrieval
- Containerized with Docker and deployed to Railway with bearer token authentication and pre-baked vector store for zero-ingestion cold starts
Multi-Agent Orchestration Platform
Agent-based research pipeline with parallel tool execution, self-critique revision, and Pydantic-validated inter-agent contracts.
- 4-agent pipeline (TopicSplitter, parallel Researchers, Synthesizer, Critic) with concurrent execution
- Self-critique revision loop evaluating factual consistency, citation accuracy, and logical coherence
- Pydantic-validated inter-agent data contracts for reliable structured data flow
Decision-Simulation Engine
Full-stack GenAI application with event sourcing, deterministic replay, scenario versioning, and admin analytics.
- Containerized full-stack app (Next.js, FastAPI, PostgreSQL, Docker Compose) with OpenAPI-documented APIs
- Event-sourced session model with custom lexer/parser/AST evaluator eliminating code injection risk
- 3,200+ line test suite and Playwright E2E coverage
Data Ingestion & Content Generation Pipeline
Multi-format document ingestion pipeline with TF-IDF content alignment and provider-agnostic LLM orchestration.
- 6-stage pipeline ingesting unstructured documents (PowerPoint, PDF, RTF, DOCX) with style-aware heading detection
- TF-IDF cosine similarity for cross-document content alignment
- Provider-agnostic LLM layer (OpenAI, Gemini, Grok) with priority-based fallback for graceful degradation
Fault-Tolerant ETL & Content Generation System
Production-grade AI pipeline with SHA256-keyed caching, rate limiting, checkpoint/resume, and multi-format export.
- 4-stage ETL pipeline with YAML-driven config, SHA256-keyed disk caching, RPM/TPM rate limiting, and exponential backoff retry
- Checkpoint/resume system eliminating redundant API calls and enabling mid-run recovery
- Multi-format export adapters (structured data API, mobile JSON API, Next.js PWA) with Pydantic v2 cross-field validators
Skills
GenAI / RAG
LangGraph, LangChain, RAG pipelines, embeddings, vector search, prompt engineering, hallucination mitigation, multi-agent orchestration
Languages & Frameworks
Python, TypeScript, SQL, FastAPI, Next.js, Pydantic v2, PyTorch, scikit-learn
Infrastructure & DevOps
AWS (7 certs), Docker, CI/CD, REST APIs, OpenAPI, structured logging, monitoring
Certifications
AI / ML
- • AWS Generative AI Developer - Professional
- • AWS Machine Learning - Specialty
- • AWS Machine Learning Engineer - Associate
- • AWS AI Practitioner
Cloud / Data
- • AWS Data Engineer - Associate
- • AWS Solutions Architect - Associate
- • AWS Cloud Practitioner