Twenty projects, all with public source

A multi-tenant AI assistant where Postgres, not the model, decides what you see

15 → 0

forbidden passages reaching the model

225

tests against a real Postgres, plus 3 live

The same assistant, sealed into a GCP network with no route to the internet

0

failed requests through upgrade, rollback and failover

24 min

to recover from a deletion, against a 60-minute target

A voice support agent that checks every sentence for grounding before it is spoken

p = 1.5e-5

McNemar exact, containment 0.737 to 0.793

0

duplicate side effects across 34 fault tests

Ad campaign dashboard over 10,000 rows, measured against a naive build

190x

faster mount at 10,000 rows

463

DOM elements, against 510,025 naive

Event-driven multi-agent AI code review on AWS

8

Lambda functions

7

SQS queues, each with its own DLQ

A distributed system built to be observed while it fails

12

scripted failure scenarios

14

alert definitions with thresholds

Agentic RAG over SEC filings, deployed and evaluated

8

specialist agents dispatched in parallel

Live

deployed on Vercel and Render

An LLM judge, and the work of validating the judge

0.41

Cohen's kappa against human labels

0.80

raw agreement with human labels

A serverless GCP platform, then a multi-agent assistant over its data

6

agents behind one supervisor

150

real CFPB mortgage complaints, queried both ways

A three-tier AWS application, provisioned entirely as code

8

subnets across 2 availability zones

5

chained security groups, ingress by SG reference

One AI endpoint, instrumented until nothing is hidden

8

custom metric instruments

11

dashboard panels, 3 comparing local to production

Snowflake-native multi-agent nutrition planning

3

medallion layers, bronze through golden

5

Cortex models compared in a judged arena

A clinical AI assistant that shows its work and its limits

4

guardrail stages before the model is called

47

patient care-gap queue

An ELT pipeline with traceable lineage into Snowflake

10

dbt models across 3 layers

20k

source tickets, multi-language

Natural language to governed SQL over BigQuery

Agentic automation with approval gates on side effects

4

Cloud Run services in one monorepo

0

irreversible actions taken without approval

LLM schema profiling behind a typed service contract

3

languages tested in one CI pipeline

1

typed endpoint contract between tiers

Student insurance platform, built by a team of five

5

engineers, team I led

3

user roles with distinct permissions

Native iOS subscription tracking with on-device inference

iOS 16+

minimum supported version

Five graduate labs on GenAI in data engineering, authored as TA

5

labs authored and maintained

26

engineers across 9 concurrent team projects