Skip to content
View ncypher's full-sized avatar
💭
1,024 uncommitted ideas
💭
1,024 uncommitted ideas

Block or report ncypher

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
ncypher/README.md

Tom Harwood / ncypher

Forward-deployed AI engineer. I sit with the people doing the work, break down what they actually need, and orchestrate teams of AI agents to ship it to production.

AI Solutions Architect at KW Corporation · Founder, CuREOsity Labs · REO Town, Lansing

CuREOsity Labs

About me / Interactive resume · Explore the collection · Enter the Journal Room · Read the philosophy

How I work

Field → Spec: on site with operators/customers; translate what they say into what they need.

Spec → Agents: decompose into staged work orders; run parallel agent lanes; I hold the context.

Agents → Evidence: nothing ships on the agents’ word — tests, runtime, and users decide.

Shipped

  • KWikSync — ServiceNow Store-certified and deployed in a large enterprise, managing a growing estate of 90,000+ IT and commercial assets, from infrastructure to edge devices.
  • Stitch
  • Q2Q middleware
  • R.E.P.O. — Vision extraction + human review.

The work behind the experiments

I build where real operations meet software: field service, warehouses, accounting, documents, integrations, and the people moving between them.

KWikSync · From the footprint to the field

KWikSync by KW Corporation

See the operation. Follow the work. Inspect the evidence.

Connected views for locations, projects, assets, surveys, and synchronization in ServiceNow. Move from the company-wide picture to the details a field team needs—and the evidence a reviewer can inspect.

Portfolio screenshots use test data. Click any image for a closer look.

See the operation · company footprint and project progress
KWikSync company footprint: a map of locations with summary cards for sites, active locations, assets, and regions. Test data.

01 · See the footprint
Locations and assets in geographic context, with the operational totals alongside them.

KWikSync projects dashboard showing active projects, ticket totals, completion status, and pending work. Test data.

02 · Follow the work
Project progress, completed tickets, and pending work in one view.

Facilities management & site surveys · proven across 2,000+ site surveys and visits

Built to support facilities management through structured site assessments, equipment documentation, photos, and human review. Proven across 2,000+ site surveys and visits, this implementation focuses on IT asset management—from infrastructure to edge devices. The survey approach can be adapted to other facility types and assessment needs.

Find the site

KWikSync locations dashboard in light theme, with a selected map marker and a View Details action. Test data.

The map connects the overall footprint to an individual location.

Review the fieldwork

KWikSync survey review showing quality feedback, assessment sections, notes, and location photos. Test data.

Survey review brings assessment notes, documentation quality, and reviewer feedback together.

The Curator · photo evidence and visual verification

Inspect the evidence

Curator photo gallery organized by site area and equipment, including network racks, security panels, and lobby views. Test data.

Organized site photos give reviewers a visual record of equipment and conditions.

Look closer with the Curator

Annotated Curator demonstration showing photo zoom, visual evidence linked to an asset record, and equipment verification. Test data.

A closer look at visual verification: zoom into equipment photos, annotate details, and relate the evidence to the asset record.

Under the hood · data synchronization and activity log

Bring the records together

KWikSync Data Sync dashboard with per-table controls for locations, projects, tickets, tasks, and assets.

A single dashboard for synchronizing operational records into ServiceNow, with per-table controls and progress indicators.

Follow the sync

KWikSync synchronization in progress, with record counts, table status, and timestamped activity messages.

The activity log makes sync progress and outcomes visible.

Stitch on a phone: the ServiceNow home screen and a field survey with pass/fail condition and leak details

Stitch
A ServiceNow app, built for the ServiceNow Store, for building field surveys, capturing evidence on a phone, and routing it for review. Shown with demo data.

cuREOsity Labs website: the what-if workbench turning a described problem into a work order

cuREOsity Labs
My REO Town technology workshop. Its site doubles as a working example: describe a headache and the workbench turns it into a work order.

System What it connects
KWikSync Field operations, locations, assets, tasks, evidence, and ServiceNow
Q2Q Accounting, time tracking, service management, fleet, and operational data
R.E.P.O. Receipts and invoices, AI extraction, validation, and human review
Warehouse management Materials, vehicles, facilities, labor, billing, and inventory history
Enterprise operations Dispatch, CRM, routing, workforce coordination, and reporting

Python · JavaScript · ServiceNow · Quickbase · REST APIs · Docker

Meet the person behind the systems → Archaeology, fieldwork, operations, and AI-assisted software: explore the career connecting them.


Start with something you can touch

"…instead of calling it work, realise it is play." — Alan Watts

Everything above is work in production. Everything below is how I keep up. When a new frontier model arrives, I give it a small, low-stakes world to build and watch what it can actually carry: a simulation, a 3D interface, a brain made of a few wires. These are toys, but the habit behind them is the same one I bring to business systems: learn exactly where the technology stands before trusting it with a warehouse, a field team, or a small business's books.

Move a control, disturb a system, and see what answers back.

Chasing the Flush: sunlight through a Michigan spring woodland at the peak of morel season

Chasing the Flush — a 3D mushroom-foraging game set in a Michigan spring woodland. Sonnet 5.5 handled the coding, testing, and deployment with Opus 5.5 orchestrating; my part was repo/Space setup and a UI walkthrough. Spring is playable; more seasons and species are planned. Explore or customize the source →

Attention Is All You Need, playable edition: two live attention diagrams, one translating an English sentence into French and one showing English words reading each other

Attention Is All You Need · Source
A Playable Paper. The 2017 Transformer paper as a working model, written and built by Claude Fable 5.2: a 7,136-number Transformer translates in your browser while you watch its attention and remove its parts.

The Chemical Basis of Morphogenesis, playable edition: a round dish covered in dark spots beside a five-lobed ring of cells and a bar chart of growth rates

The Chemical Basis of Morphogenesis · The shelf
A Playable Paper. Turing’s 1952 dish of two chemicals, also by Fable 5.2: it grows spots, stripes, and nets on its own until you take away the difference in how fast they spread.

Claude Opus 5.5's 8.5 Ball: a black 3D ball with a blue twenty-sided die answering a question, and its reasoning trace beside it

8.5 Ball × 2 · ChatGPT’s
A Magic 8 Ball for the age of AI, built by ChatGPT, then rebuilt by Claude Opus 5.5 as a 3D oracle that shows its reasoning and admits it can’t see the future.

Claude Opus 5.5's Rubik's Cube mid-solve, with the stage checklist and solution

Rubik’s Cube × 3 · GPT-5.6 Sol’s
One puzzle, three generations. In Opus 5.5’s, the layers really turn and the solver finishes from any position instead of rewinding your moves.

Intelligence Explosion: a tree of AI-assisted discoveries branching across biology, mathematics, and physics

Intelligence Explosion
Is discovery accelerating? A sourced, living atlas of AI-assisted breakthroughs, with what the AI did and where the evidence stops.

Gold Digger: the S&P 500 re-priced in gold, with a chart comparing dollar and gold returns

Gold Digger
Re-price stocks, funds, and crypto in gold, eggs, gasoline, or houses to see what an investment is really doing.

Claude Opus 5.5's Pixel Pet: a grown-up pixel creature leaping for a ball in its room at evening

Pixel Pet × 2 · The original
Hatch a one-of-a-kind creature and raise it in a room on your real clock; how you care for it decides what it grows into.

Crucible: a low-poly 3D island with pine trees, boulders and a lone survivor by a campfire, beside sliders for weather, scarcity and danger

Crucible Dynamic
Wind a little island world, set its weather, scarcity, and danger, then let go and watch who the stranger on the shore becomes over seven days.

Wind-Up Minds: Braitenberg carts circling two lights, a worm finding food beside its live neuron circuit, and a tiny network's learned boundary between orange and blue dots

Wind-Up Minds · Vehicles · Worm · Network
Three small brains you can poke: two-wire carts that flee, charge, or fall in love with light; a worm run by 21 real C. elegans neurons; and a from-scratch Python network learning live in your browser.

More to explore Follow the question
Polycentric Orchestra What changes when evidence and human acceptance govern the loop?
Feedback Knot What happens when development doubles back on itself?
What the Wren Hears Can you understand a radio signal by playing with its shape?

Streamlit worlds: Gold Digger · Crucible Dynamic · Tiny Gods · The Turning · AI Think Tank · Synapse · Office Hours.

Community work: Project reoWren explores community-owned reporting and radio-mesh communication. Follow its proof-of-concept development on Patreon.

The early browser games are still here. Their code was entirely generated by the frontier models available when I made them; I supplied prompts, guidance, and testing. I was the crash test dummy. They preserve what those models could make, the questions I was asking, and the revisions that followed.

The practice behind it

A human-directed feedback loop: intent becomes a proposal, implementation meets review, dissent triggers revision, and runtime evidence returns for provisional human acceptance.

The diagram follows a scored feedback cycle. Click it to introduce dissent, fail a test, and decide when the work is ready.

How I work

The conversation is often the first draft. An observation becomes a question; the question becomes a sketch; the sketch becomes something we can test.

I work with Claude, Codex, Gemini, NotebookLM, Grok, and tools for images, music, research, and implementation. Roles change with the task. A useful collaborator can propose, build, question, or reveal something the others missed.

The responsibilities stay distinct:

  • Human direction: purpose, boundaries, taste, and acceptance.
  • Generation and criticism: make a proposal, expose assumptions, preserve disagreement.
  • Verification: compare the claim with tests, runtime behavior, and user experience.
  • Revision: let what happened change what happens next.

Confidence is not evidence. Passing a test does not transfer responsibility to the test. An accepted result can still be reopened when new evidence arrives.

Software is not the only product here. Understanding is.

I am learning the formal language of cybernetics through patterns I first met in practice: feedback, distributed judgment, delayed consequences, and systems that can correct their own mistakes. This is a developing practice, with a notebook to match.

Philosophy · Workflow · Timeline

Still observing

New field notes · September 30, 2026: Repo Excavations II — When the Map Has to Answer to the Territory follows aging evidence, shared meanings, and Ghost’s archived layers. I Was the Crash Test Dummy records what the model-generated games were exploring.

Can a system preserve a useful minority report? Can an interface teach an idea before naming it? Can environmental sensing help a neighborhood notice its own feedback?

The Journal Room holds the observations, contradictions, and field reports behind the artifacts. Entries can remain a paragraph, become software, or change my mind and stop there.


Do not make yourself write. Make yourself wonder.

About Tom · Conversational Artifacts · GitHub · X

Popular repositories Loading

  1. ncypher ncypher Public

    JavaScript 3 2

  2. KW-DataSync KW-DataSync Public

  3. tomfoolery tomfoolery Public

    a repo for fun little vibe coded games

    HTML

  4. -Claude-Memory-System -Claude-Memory-System Public template

    🧠⚡ Transform Claude into your persistent AI partner! No setup, no external storage - just conversation-search magic for true AI memory continuity.

  5. KWikSync-docs KWikSync-docs Public

    Documentation hub for KWikSync, a ServiceNow Store application that enables seamless data synchronization between ServiceNow and external systems via IntegrationHub ETL.

  6. gold_digger gold_digger Public

    Python