Atlassian
2023–Present- Principal Data Scientist2025–Present
- Senior Data Scientist2023–2025
Applied AI for customer support and knowledge creation. Combining prototypes, experimentation, and technical direction to improve how customers get help.
Applied AI & Machine Learning
JoeDeGregorio.
I find valuable problems and build the AI & ML systems that solve them.
Business judgment, modeling depth, and hands-on technical leadership. From the first working prototype to a measurable difference in how a business operates.
01 / Selected work
Useful systems. Measurable change.
Selected work in workflow automation, predictive modeling, and customer support. Explore the problem, my contribution, and what changed.
Turning support knowledge into an AI workflow
An agentic system that researches, drafts, fact-checks, and revises support articles—helping teams close knowledge gaps and customers find answers.
Multi-agent systems · Grounding · Quality evaluation
Read the storyClose the story~80%less article creation time
Turning support knowledge into an AI workflow
An agentic system that researches, drafts, fact-checks, and revises support articles—helping teams close knowledge gaps and customers find answers.
Multi-agent systems · Grounding · Quality evaluation
Read the storyClose the storyThe opportunity
Support knowledge creation was manual and reactive. Articles lagged behind customer needs, limiting both self-service and the AI support experiences that depended on them. The opportunity was to improve the whole knowledge workflow, from finding gaps to producing useful, grounded content.
My contribution
Designed a pipeline with specialized agents for research, drafting, factual verification, and quality review. Built review loops around source grounding and publishing standards, then extended the system into reusable capabilities for gap detection, generation, revision, and evaluation, using classical ML and LLMs where each fit.
What changed
Reduced article creation time by roughly 80%. A pilot batch of AI-generated articles began deflecting tickets in production, and adjacent product teams adopted the underlying Knowledge AI capabilities. The work helped shape their roadmaps around proactive knowledge creation.
Predicting failures. Prioritizing the right repairs.
Survival models that identified at-risk engines and directed limited dealer capacity toward repairs with the greatest potential to prevent costly failures.
Survival analysis · Simulation · Capacity-aware prioritization
Read the storyClose the story$13Min warranty costs saved
Predicting failures. Prioritizing the right repairs.
Survival models that identified at-risk engines and directed limited dealer capacity toward repairs with the greatest potential to prevent costly failures.
Survival analysis · Simulation · Capacity-aware prioritization
Read the storyClose the storyThe opportunity
An engine defect created roughly $40M in projected annual warranty exposure. Preventive rebuilds cost $2,500; catastrophic failures could exceed $10,000. With limited dealership capacity, the business needed to decide which engines to bring in first.
My contribution
Led a ten-person team across data engineering, ML, and warranty analysis. Built Cox survival models to estimate each engine’s 30-day failure risk from telematics, plus usage simulations to forecast mileage and engine hours. The outputs drove monthly recall priorities, engineering root-cause reviews, and financial reserve forecasts, with weekly reporting to the CEO and CFO.
What changed
Saved $13M in warranty costs by prioritizing preventive repairs before catastrophic failures. The program reduced unplanned customer downtime, made better use of constrained dealer capacity, and gave finance a defensible basis for warranty reserves.
A better path from customer question to resolution
Connected self-help, search, live chat, routing, and AI support through a shared measurement foundation to improve how customers get help.
Experimentation · Product strategy · Impact measurement
Read the storyClose the story$7M+in documented cost avoidance
A better path from customer question to resolution
Connected self-help, search, live chat, routing, and AI support through a shared measurement foundation to improve how customers get help.
Experimentation · Product strategy · Impact measurement
Read the storyClose the storyThe opportunity
Customer support spanned disconnected experiences with real cost and customer stakes. Teams needed a way to identify the most useful interventions and understand whether changes actually improved resolution and avoided support costs.
My contribution
Established instrumentation and experimentation across the support journey, then used that evidence to guide changes to live chat routing, search, self-help, and AI answers. Paired product decisions with measurements of customer experience and cost avoidance.
What changed
The multi-year program delivered $7M+ in documented cost avoidance alongside customer-experience gains. One routing improvement increased the share routed to live chat from 19% to 35%, with a material reduction in time to resolution.
Formloop: from design intent to 3D geometry
A hands-on exploration of agentic engineering: natural-language design intent becomes CAD geometry through tool use, specialist agents, and inspectable review loops.
Python · Agent orchestration · CAD tools
Read the storyClose the storyOpen sourceworking code, artifacts, and review traces
Formloop: from design intent to 3D geometry
A hands-on exploration of agentic engineering: natural-language design intent becomes CAD geometry through tool use, specialist agents, and inspectable review loops.
Python · Agent orchestration · CAD tools
Read the storyClose the storyThe opportunity
Agentic engineering is easier to assess with working artifacts. I built Formloop to explore how agents can coordinate around real engineering tools and make their design decisions and quality checks visible.
My contribution
Built a Python harness coordinating a manager agent and specialists for CAD design, review, judging, narration, and research around a deterministic toolchain. Runs produce CAD code, STEP and GLB files, renders, progress events, and review traces.
What changed
An open-source reference implementation with code and artifacts that make the orchestration inspectable. Built as a personal prototype for exploring and evaluating tool-grounded AI workflows.
Explore Formloop on GitHub02 / How I work
Find the opportunity. Make it real.
I connect problem discovery, quantitative reasoning, and building—keeping the business outcome in view at every step.
Find the right problem
Get close to the workflow, investigate the root cause, and size the opportunity. Turn a broad business objective into a specific problem worth solving.
Model the decision
Connect predictions to actions, costs, and constraints. Choose the approach that fits, whether that is a heuristic, a statistical model, or an AI workflow.
Build to learn
Make the idea tangible quickly. Use working prototypes to test assumptions with users, expose failure modes, and sharpen product and technical requirements.
Follow through to value
Partner with product, engineering, and domain experts to put the solution to work. Measure quality, adoption, and business outcomes to guide the next investment.
03 / Experience
Engineering roots. Applied AI reach.
I started in mechanical and systems engineering, then moved into data science and applied AI. That path helps me connect how a business works, what the data says, and what a useful system needs to do.
PACCAR
2016–2023- Senior Analyst → Data Scientist2018–2023
- Senior Systems Engineer2016–2018
Predictive models and decision systems for reliability, warranty, and operations. Connecting engineering context and statistical reasoning to decisions with real financial stakes.
Caterpillar
2012–2016- Senior Design Engineer2012–2016
Complex mechanical design and root-cause analysis. The foundation for how I investigate problems, reason about failure, and design around real-world constraints.
04 / Let’s talk
Have a problem
worth solving?
Let’s talk about applied AI, intelligent workflows, and new products that make a measurable difference to a business and the people doing the work.