Projects

Applied AI — on top of a decade of delivery.

Two threads of work: the AI tooling I’m building now at the intersection of LLMs, data engineering, and self-serve analytics — and the foundation of shipped governance, forecasting, and team-building that came before it.

Note Much of my current work at Verizon is under NDA. The case studies below are anonymized methodology write-ups of patterns I’ve shipped in production — no proprietary metrics, customer data, or internal system names.

Applied AI · current focus

LLMs in production for data teams.

Conversational BI for leadership

shipped

Natural-language interface over a governed LookML semantic layer so leadership can query BigQuery without writing SQL. Guardrails, citations back to source models, and an eval harness for accuracy.

<5 min
question → answer
#gemini#claude#rag#bigquery#looker

Analyst Copilot in the reporting suite

shipped

AI assistant embedded in our Looker workflow that drafts narratives for weekly readouts, summarizes anomalies, and answers stakeholder follow-ups in context.

~6 hrs/wk
analyst time recovered
#gemini#looker#automation

Teradata → BigQuery migration assistant

in-progress

LLM-assisted translator for legacy Teradata SQL into BigQuery models, paired with a deterministic diff harness that validates the new model produces identical rows before cutover.

120+
models migrated
#teradata#bigquery#claude#migration

AI-assisted model docs & lineage

in-progress

Auto-generates documentation for BigQuery models and LookML, suggests data checks, and flags downstream impact when a schema changes &mdash; closing the doc gap that always lags model velocity.

90%+
model docs coverage
#bigquery#lookml#claude#data-engineering

KPI anomaly explainer

in-progress

When a headline KPI moves, an LLM walks the pre-computed segment breakdowns and drafts a plain-English explanation &mdash; turning &ldquo;something&rsquo;s weird&rdquo; into a starting hypothesis.

#gemini#looker#anomaly-detection

Eval harness for conversational analytics

concept

Methodology write-up on how to evaluate a &ldquo;talk to your data&rdquo; product &mdash; gold-question sets, semantic similarity scoring, and the failure modes that look correct.

#evaluation#methodology#case-study

Selected delivery · Verizon

Governance, forecasting, and the team behind it.