Solutions Architect · Generative AI · Cloud Platforms
Senior Solutions Architect in AWS Worldwide Public Sector. Nineteen years across software engineering, infrastructure, and cloud architecture, the last twelve in education. Current focus is generative and agentic AI, serverless architecture, and ML data platforms for education technology.
Curtis Darst is a Senior Solutions Architect in AWS Worldwide Public Sector, where he owns the technical relationship for a portfolio of strategic EdTech ISVs worth $38.5M in annual recurring revenue. Their products cover learning management, student information systems, assessment platforms, enrollment management, and adaptive courseware. They reach K-12 districts, universities, and state education agencies.
That work covers generative AI and agentic architectures on Amazon Bedrock, on-premises to cloud migrations, serverless modernization, and the data foundations ML workloads need. Before the strategic accounts role he was lead technical advisor to research universities and liberal arts colleges across a Great Lakes and NJ/PA territory.
Before AWS, he built statewide K-12 data, assessment, and federal reporting systems for the West Virginia Department of Education, directed engineering and server infrastructure at a market research firm, and taught mathematics as an adjunct instructor. He is a Ph.D. candidate in Computer Science researching how prompt engineering patterns affect the quality of LLM-generated code, and lives in Powell, Ohio, just north of The Ohio State University campus.
Nineteen years spanning statewide public-sector systems, in-house engineering leadership, and customer-facing cloud architecture.
Amazon Web Services, Worldwide Public Sector · Remote, Powell, OH
Amazon Web Services, Worldwide Public Sector · Remote, Powell, OH
Amazon Web Services, Worldwide Public Sector · Remote
Across all three roles: Technical Field Community member for Serverless, Serverless Compute + APIs, and Application Integration. AI-Driven Development Lifecycle (AI-DLC) Ambassador. L400 focus-area recognition in Generative AI and Amazon MWAA; L300 competencies in Blog Author and Mentoring. Currently mentoring 10 solutions architects.
West Virginia Department of Education · Charleston, WV
West Virginia Department of Education · Charleston, WV
Developed software applications supporting statewide education data systems in ColdFusion and related toolsets. Promoted to Coordinator after 16 months.
McMillion Research · Charleston, WV
Mountwest Community & Technical College / Marshall CTC · Huntington, WV
Taught mathematics to community and technical college students, concurrent with full-time work at McMillion Research.
Benchmarks, prototypes, and practitioner guidance, most of it built to answer a question a customer asked.
Exports one model to four formats and deploys it across nine SageMaker AI configurations covering CPU, GPU, Inferentia2, and Serverless. Measures P50/P95/P99 latency, throughput under concurrency, and cold start, then reports cost per million inferences. Cost varies by more than 40× across configurations serving the identical model. Includes automated teardown with verified deletion, a runtime cost gate, and 166 unit tests.
A multi-agent system built with the Strands Agents SDK and Amazon Bedrock AgentCore, taken from whiteboard through working prototype and hand-off. Covers agentic patterns, retrieval, evaluations, and guardrails for workloads that handle student data.
A high-volume serverless pipeline for processing and delivering instruction files across development, QA, and production environments, built for horizontal scale, fault tolerance, and cost and performance efficiency at each stage.
Guidance for enterprise leaders on data-flow control and vetting risk when installing Model Context Protocol servers from public repositories. Written for the teams deciding what gets into their developer toolchain.
Everything public on github.com/curtisdarst, newest first. The list refreshes from the GitHub API when the page loads.
Showing the last saved snapshot
ADK research triage agent on Google Cloud that verifies every claim against its retrieved source and reports what it cannot support. Ships with a reproducible citation-grounding eval.
A Minesweeper-clone testbed for local LLMs. Tiered tools isolate perception from reasoning, an exact solver grades every move against what was provable, and models can induce their own rules through a verifier gate. Runs on 8GB VRAM.
Reproducible benchmarking framework for ModernBERT inference on Amazon SageMaker AI. Deploys one model across CPU, GPU, Inferentia2, and Serverless endpoints, measures latency, throughput, batching, and cold start, then reports cost per million inferences and a weighted decision matrix. Includes measured results for nine configurations.
AWS Workspaces Automation
Reviewing the context window limits for Bedrock Nova
CodeMash 2022 Session
Raleigh Code Camp 2019: SERVERLESS COMPUTE WITH AWS AND .NET CORE
Event: Raleigh Code Camp 2019 Session: Serverless Compute with AWS and .NET Core
Using AWS Transcribe service, take the JSON output and output HTML into a nice conversational format for easy posting to a blog.
Raspberry Jam Talk on 03022019: From Click to Cluck: IOT interactions with a Raspberry Pi and AWS
Raspberry PI Door Sensors
Candidate · Expected Dec 2027
Dissertation: The Effect of Prompt Engineering Patterns on LLM-Generated Code Quality. A quantitative quasi-experimental repeated-measures study producing 525 code samples across seven models, five prompt patterns, and fifteen Python scenarios. It tests whether prompting technique measurably affects weighted defect density and cyclomatic complexity, and whether that effect differs between traditional single-pass models and extended-thinking reasoning architectures.
Cumulative doctoral GPA 3.90 across 39 credits.
Marshall University · 2013
Comprehensive project: an Augmentative and Alternative Communication (AAC) iPad application for users with speech and language disabilities, spanning both software development and project management.
Marshall University · 2007
Senior project: FARM Assessment Resources Management System, an application for tracking livestock, income, and expenses that produced algorithm-driven planning output for farm operations.
AWS Tier 2 Certified Speaker and published AWS blog author. Presenter at AWS re:Invent, re:Inforce, and four AWS Summits, across serverless, security, and developer experience.
Eight active AWS certifications across the foundational, associate, and professional tiers.
AWS Certifications
Accreditations & Recognition
For architecture conversations, speaking invitations, mentoring, or research collaboration: