Hailey Cummings

Oh, you actually looked me up.. Hi!

Hailey Cummings

Project manager who builds with AI.

The plan

  • Eleven years delivering enterprise programs
  • Agile, waterfall, hybrid: whatever fits the work
  • ADKAR-led change that sticks

The build

  • Python projects, shipped in public
  • AI tooling adoption, from pilot to habit
  • IBM AI Engineering certificate, completed

 

See the work Talk to me

Trusted by teams at Chevron, Electronic Arts, HPE, and Sysco.

Credentials

Titles vary. The work doesn't.

Job boards file this work under program manager, change manager, AI program lead, technical program manager, and forward-deployed or solutions roles. If you're matching on skills instead of titles, here is where I add the most value.

  1. When the problem is still fuzzy. I sit with executives and customers until the real need is clear, then scope it and de-risk it.
  2. Between the business and the builders. I translate in both directions, so engineers build the right thing and sponsors know what they're getting.
  3. When adoption is the point. ADKAR-based change planning turns a launch into a habit, which is where most AI tooling programs stall.
  4. When the question is which AI use case comes first. I rank the candidates against leadership goals, build the business case, and set the stop-gates that keep a pilot honest.
  5. When I need to prototype it myself. Python and AI tools let me put something in front of people early, instead of describing it.

Skills, with receipts

A list of strengths is easy to write. Here is what each one looked like on real work.

Project management
Eleven years running enterprise programs from intake to handoff, in agile, waterfall, and hybrid shops. I choose the method that fits the work, then keep scope, schedule, and risk visible. I directed 15+ initiatives worth $5M+ each at 95%+ on-time delivery, and stood up a PMO for 10 AI workstreams that cut weekly status prep from four hours to one and closed 27 of 35 logged risks in 30 days.
Stakeholder management
Built working relationships with C-suite sponsors, Fortune 500 clients, and cross-functional teams, so hard tradeoffs get decided early instead of escalating late.
Change management
ADKAR-based adoption plans that treat people as the deliverable. I led an AI program at a large global software company that grew from 500 to 10,000 users in six months at 75% adoption, counting only people actively using it at least three times a week.
Delivery and value
Turned requests into numbers: a self-service portal that cut fulfillment time by 73%, an online assessment platform for learning, and a value-creation plan for a large manufacturer.
Customer-facing delivery
Helped scope nine client proposals, drafting scope and resource estimates behind $3M in pursued work. Ran offshore handoffs across three time zones without letting the on-time rate slip, and led implementation plans across 147 monitored accounts that held churn to five against an expected eight to twelve.
AI use case prioritization
Identified and ranked six priority AI use cases across five business units, then tied the rollout order to C-suite goals so the first launches were the ones leaders cared about.
Business case and value tracking
Builds the case, then tracks the benefit. Avoided $1.2M in rollout costs over seven months by surfacing overlooked risks and re-prioritizing use cases, and documented a workflow automation's business case so others could reuse it.
Decision frameworks and stop-gates
Built the framework for an agentic AI pilot under client security rules: phased ADKAR plans, rollout qualification criteria, and reusable stop-gates. Three use cases went from pilot to production-ready in four months.
Responsible AI governance
Partnered with Legal on AI governance and resolved nine policy questions, so teams could move without guessing what was allowed.
AI enablement
I help organizations choose AI tools, set sensible guardrails, and teach people to use them well. I know how these tools work and where they break, which keeps the adoption plan honest.
Data, Python, and AI engineering
I build with Python, SQL, and Power BI, and I automate recurring workflows (about five hours saved a month, documented for reuse). I completed IBM's AI Engineering and RAG and Agentic AI certificates, and my projects below are how I'm applying them.

Daily tools: Jira, Confluence, Salesforce, Smartsheet, SharePoint, Teams, PowerPoint, Excel. Python, SQL, and Power BI as I build.

The forward-deployed toolkit

The tools forward-deployed and AI engineers reach for, stage by stage, with the spot where project management earns its keep. Filled dots are tools I've used in my own work, half-filled are ones I've studied, and hollow ones are on my map.

  1. Discover
  2. Analyze
  3. Connect
  4. Build
  5. Ground
  6. Evaluate
  7. Ship and adopt

Discover

Find the real problem before anyone writes code. Forward-deployed work starts in the customer's world: their data, their workflow, and the person who feels the pain.

  • Jira tickets and delivery tracking (In my work)
  • Confluence discovery notes and decisions (In my work)
  • Smartsheet plans and resourcing (In my work)
  • SQL query the customer's data (In my work)
  • Salesforce accounts and pipeline (In my work)
  • Jupyter explore data fast (On my map)
  • Postman poke at customer APIs (On my map)

Where the plan matters. Stakeholder interviews, a success metric, and a risk list come before the first prototype.

Analyze

Turn the customer's data into a decision. Pick the use cases worth building, size the benefit, and show leaders the numbers.

  • Power BI dashboards for leaders (In my work)
  • Python analysis and automation (In my work)
  • Hadoop and Spark three years of big data work (In my work)
  • R and scikit-learn IBM coursework (Studied)
  • Tableau visual analytics (In my work)

Where the plan matters. Rank use cases by value and effort against leadership goals, and write down what "worth it" means before the build.

Connect

Wire the model into the systems people already use. Most of the effort here is access, permissions, and plumbing, not AI.

  • Composio Gmail, Outlook, and Sheets connections (In my work)
  • MCP standard way to give models tools (On my map)
  • REST and webhooks move data between systems (On my map)
  • n8n and Zapier low-code glue (On my map)
  • OAuth and API keys access done safely (On my map)

Where the plan matters. Security review and data access are schedule risks. Name them in week one, with owners.

Build

Ship a first version that does one useful thing. Agents and chains come later, once the simple version has proved its worth.

  • Python my projects, end to end (In my work)
  • Claude assistant and coding partner (In my work)
  • ChatGPT and Copilot assistants I help teams adopt (In my work)
  • LLM APIs Anthropic, OpenAI, and others (On my map)
  • LangChain and LangGraph chains and agent workflows (On my map)
  • FastAPI serve a model behind an endpoint (On my map)
  • PyTorch and TensorFlow IBM coursework (Studied)
  • Hugging Face open models and datasets (On my map)

Where the plan matters. Scope the smallest slice that a real user would use, then protect it from additions.

Ground

Make answers come from the customer's own sources instead of the model's memory. Retrieval is where most enterprise assistants succeed or fail.

  • RAG retrieval-augmented generation (On my map)
  • Embeddings turn text into searchable vectors (On my map)
  • Chroma and FAISS local vector search (On my map)
  • pgvector and Pinecone vector search at scale (On my map)

Where the plan matters. Agree on the source of truth and who keeps it current. Stale documents make confident wrong answers.

Evaluate

Prove it works before anyone relies on it. A small test set and a clear pass mark beat a good demo.

  • Promptfoo and Ragas test prompts and retrieval (On my map)
  • LangSmith and Langfuse trace what the model did (On my map)
  • pytest test the code around the model (On my map)
  • Review rubric human acceptance checks (In my work)

Where the plan matters. Acceptance criteria, a go or no-go review, and a named owner for every failure.

Ship and adopt

Release it, then get it used. A launch that nobody changes a habit for is not finished.

  • GitHub version control and review (In my work)
  • Cloudflare Workers this site runs here (In my work)
  • ADKAR the adoption method I lead with (In my work)
  • Docker package it to run anywhere (On my map)
  • GitHub Actions automate tests and deploys (On my map)
  • AWS, Azure, and GCP the clouds customers run on (Studied)

Where the plan matters. Training, champions, and a feedback loop turn a release into a habit.

Things I'm building

PM Arsenal

Four tools a project manager reaches for every week, in one place: a Risk Scanner, an Executive Briefing Generator, a Capacity Tracker, and an AI Scenario Roleplay for practicing stakeholder situations.

In progress. I'm finishing the build on GitHub, then it goes live here.

See the code on GitHub

A private place to document abuse, and a person to sit beside you in court

Next build. Shown as a case study for now: a concept, not a live service. Any demo uses made-up data.

For people leaving abusive situations who aren't ready to call the police. They log incidents with dates and times, assemble them into a packet for an investigator, get plain-language answers about the legal process, find free local support, and are matched with a trained volunteer to attend court with them.

  • Built for a shared or watched device. A quick-exit button, a neutral name and icon, and notifications that never reveal the purpose.
  • A log an investigator can trust. Entries are timestamped by the server, additions are allowed, and original entries keep their history.
  • A chatbot that stays in its lane. It explains the process from vetted sources, does not give legal advice, and surfaces hotline information whenever someone describes danger.
  • Volunteers as a safety program. Vetting, training, and a way to report problems come first, ideally with an existing advocacy organization.
  • Store less. Retention and what a court could compel are decided up front.
Proposed architecture
  1. Neutral appPlain name and icon, quick-exit button
  2. Private incident logServer timestamps, append-only history
  3. Evidence packetAssembled for an investigator
  4. Process assistantRetrieval over vetted sources, hotline shown when danger is described
  5. Court buddiesVetted, trained volunteers, matched by request

If you need help now: National Domestic Violence Hotline, 1-800-799-7233 or thehotline.org.

Job-search pipeline automation

Connected Gmail, Outlook, and Google Sheets so applications, follow-ups, and fractional leads land in one tracker without copy and paste. A small build that shows how I think about workflows: find the manual step, then remove it.

Set up for my own search. A write-up is coming.

How it's wired
  1. Gmail and OutlookWhere applications and replies arrive
  2. ComposioConnects the inboxes and the sheet
  3. Google Sheets trackerApplications, follow-ups, and leads in one place

Also in the works

  • Trip 33. A group trip coordination app with energy-aware itinerary pacing. The MVP spec, constraint taxonomy, and data schema are done, and version one is scoped around the Duffel, Viator, Booking.com, and Google Places APIs.
  • Kids' digital safety app and course. Phone etiquette and online safety for children aged seven to sixteen, because kids get a device early and rarely get a plain-language talk about what safe use looks like. Early stage.

Now building, in public

Real repositories, listed by most recent update, straight from GitHub.

How this site is built

  1. Static pagesHTML, CSS, and a little JavaScript. No framework, self-hosted fonts
  2. Cloudflare WorkersServes the site from the edge worldwide
  3. Push to GitHubEvery change to main deploys automatically
  4. An assistant, readyA Worker endpoint for a live Q&A chat is built and switched off for now

Common questions

What recruiters and hiring managers ask most, answered straight.

How does a project management background help in a forward-deployed role?

Forward-deployed work starts with a customer whose problem is only half described. I'm used to sitting with executives and operators, finding the real need, scoping it, and managing the risk. Then I can prototype in Python and AI tools, and get the people involved to use what we built. The combination of customer trust, delivery discipline, and technical fluency is the job.

Which companies have you supported?

Teams at Chevron, Electronic Arts, HPE, and Sysco have trusted me with their programs. I describe the work in general terms and don't tie any project, result, or tool to a specific client.

Which tools do you use, and which are you still learning?

In my own work: Python, SQL, Power BI, Claude, ChatGPT and Copilot, Composio, Jira, Confluence, Salesforce, Smartsheet, GitHub, Cloudflare Workers, SharePoint, Teams, PowerPoint, and Excel, plus ADKAR and human review rubrics for acceptance checks. Studied through IBM coursework: R, scikit-learn, PyTorch, TensorFlow, and the major clouds. On my learning map, not claimed as production experience: LangChain and LangGraph, FastAPI, RAG patterns and vector stores, evaluation tools like Promptfoo and Ragas, Docker, and GitHub Actions. The toolkit section marks each one.

Why you for an AI program lead role?

I spent a year leading an AI program that grew from 500 to 10,000 users in six months at 75% adoption, and avoided $1.2M in rollout costs along the way. I ranked the use cases, built the stop-gates for an agentic pilot, and worked with Legal on governance. That is the hard part of most AI programs, and it's a people problem more than a model problem. I also build, so I can judge scope and risk with engineers instead of relying on summaries.

How deep is your hands-on AI engineering?

Newer than my delivery and change experience, and I'd rather say so than hide it. I completed the IBM AI Engineering Professional Certificate and the IBM RAG and Agentic AI Professional Certificate, and I build projects in public so you can judge how I learn from the code, not a claim.

What's your path in short?

Eleven years in enterprise project and program delivery, growing into change management for Fortune 500 clients, then fractional consulting since August 2024. I most recently led an AI tooling adoption program at a large global software company, and I'm now moving toward director-level and hybrid PM and AI engineering roles.

Field notes

Adoption is the deliverable. A tool that nobody changes a habit for is shelfware. ADKAR keeps the plan honest about awareness, desire, knowledge, ability, and reinforcement, in that order.

Ask what the requester sees. A self-service path beat a faster team, because it removed the waiting instead of speeding it up.

Make risk boring. When risks are listed, owned, and reviewed on a schedule, they stop being surprises and start being line items.

Hiring for a role that sits between the plan and the build?

haileyscummings@outlook.com

LinkedIn

I'm looking at director-level and hybrid PM and AI engineering roles, including forward-deployed work.