Hey — I'm Adam Crocker. AI Cloud Engineer. AWS Solutions Architect — Professional (Jan 2025, valid Jan 2028). Based in Springville, Alabama; remote anywhere.
I build AI-native systems for businesses of every size — Fortune 500 down to mom-and-pop shops. Same person, same care. Most of my work sits at the intersection of LLMs & agents, AWS, and production data engineering — and the integration work between them.
Here are six pieces I'd want a hiring engineer to look at — different shapes, all production:
Lead engineer on the central operational data warehouse. TimescaleDB with 15 schemas, 77 tables, 114 PL/pgSQL functions, 1.6M+ daily production records. Star schema, hypertables, continuous aggregates. ETL from CygNet SCADA, Snowflake, MNR; outbound to Grafana and regulatory reporting. Embedded pgai for automated issue detection over operator free-text and semantic search across historical observations.
10-agent system handling the full client lifecycle — intake, sales, scoping, build, QA, deploy, support, billing, comms, phone — across 8 project types. MiniMax M2.7 primary; Gemini 3 Flash Preview for visual QA. Voice via Pipecat + Telnyx + Deepgram Nova; agents execute inside Cua sandboxes. SvelteKit client portal on Vercel.
JSON-first replacement for three legacy scorecards (~2,300 entities/closeout). Python packagers on Teradata, Step Functions parallel branch, DynamoDB caching, single vanilla-JS viewer. AWS Bedrock (Claude Haiku 4.5) for auto-narratives + per-KPI explanations via /api/v1/kpi-explain. Closed seven latent defects en route, including an eval()-based templating injection.
Want a closer look at any of these? Or there are three more — .
The other three:
ML platform over 186M+ telemetry observations. 8 categories of industrial sensor failure. Isolation Forest for unsupervised anomaly detection, Random Forest for pattern classification, LSTM for sensor-level forecasting. Packaged as an AWS CDK stack with a 580-line handoff runbook.
5-stage Azure DevOps pipeline on a self-hosted Windows agent. SQL risk classifier (BLOCKED / HIGH / MEDIUM / LOW) over ~30 regex patterns. DROP TABLE foreign-key safeguard. Transactional execution with auto-rollback. SOX-aligned audit logging.
Ionic 8 + Angular 20 + Capacitor. Single codebase → iOS, Android, installable PWA. Offline-first on Dexie / IndexedDB with a write-queue and conflict-resolving sync. AWS AppSync + DynamoDB; Cognito federated to Azure AD via MSAL.
I've been at Delta since 2021. Two phases:
2021–23: Associate Software Developer. The early-career half — Executive AI Reporting POC, AWS Glue CI/CD work, STS SSO automation pure-bash (worked around Windows Defender + admin restrictions on the corporate fleet).
2023–present: Data Engineer on the Operations Strategy Reporting (ODSR) team. The work I'm most known for internally: Reporting Modernization v2. The SCRD / ESR / MPR scorecards were huge PPTX / HTML monoliths — I led the replacement onto a JSON-first stack on AWS-native services (Glue, Step Functions, DynamoDB, S3, API Gateway, Bedrock), with Claude Haiku 4.5 generating auto-narratives and per-KPI explanations.
The constraint was real: delegate-admin- IAM boundary plus an SCP blocking new Lambda creation. I shipped the AI compute as a Glue Python-shell job with custom boto3 packaging so we didn't need an architecture exception.
June 2026: selected for Delta's AI Enablement Team. AI agent standards across Operations Planning & Performance. That's the next chapter.
SCRD = Scorecard. It's an internal Delta operational scorecard. The v1 ran as a generated PPTX with per-business-unit slides; v2 is a JSON-first data package + a single vanilla-JS Chart.js viewer. The win wasn't just the renderer — it was making one packager handle ~2,300 organizational entities per monthly close, with the renderer code surface shrinking enough that Python-heavy teammates could maintain the frontend.
ESR (Executive Strategy Reporting) and MPR (Monthly Performance Reporting) ride the same pipeline now.
While reviewing the legacy SCRD code I found a template renderer using eval() on operator-supplied strings — a templating-injection vector hiding in plain sight. Closed it as part of the v2 migration along with six other latent defects (the worst was swapped year/month columns on one of the rollups, silently miscategorizing closeouts).
Total: seven defects closed during the modernization, all logged with reproduction steps in the 13 engineering docs.
I joined Diversified Gas & Oil as a Systems Engineer in August 2025, concurrent with Delta. Three core systems I own:
Smaller but adjacent: Hybrid DB Migration CI/CD — a 5-stage Azure DevOps pipeline that classifies every schema change against ~30 regex risk patterns and gates accordingly. SOX-aligned audit logging.
That Simple Tech (TST) is my AI-native technology consultancy. Sole proprietor / operator. The "headline product" is the Autonomous AI Agency Platform — a 10-agent system that handles the full client lifecycle for small-business engagements:
intake · sales · scoping · build · QA · deploy · support · billing · comms · phone
Architecture decisions worth flagging:
Client portals never expose AI terminology — to clients it just looks like a really fast agency. There's a VA review gate (PENDING_REVIEW) on all client-facing output.
Six buckets, sixty-plus tools. Ask me about anything specific.
Easiest paths in:
Springville · Alabama · US. CT time zone, remote anywhere.
Organizations I've done work with or for:
The work itself — projects, scale, decisions made — is the better story. ↗
That's a personal one — I'd rather not get into the employment math here. The portfolio is about the work itself. Happy to dig into any specific project though.
Good question — but I haven't been pre-authored to answer that one. This portfolio is static; the responses you see are written by the human Adam, not generated. Try one of:
Or just email me.