Calm fronts
Public pages stay clear, restrained, and useful.
The visible layer should feel trustworthy and easy to navigate, even when the machinery behind it is much more powerful.
Live portfolio / useful systems
Alex Harper
I build systems that build bigger systems: domain research, operator fabrics, AI-assisted workflows, browser tools, embodied 3D experiments, mobile companions, and the connective tissue that keeps them useful after launch.
Media intelligence
Audio, speech, story, and continuity systems
Operator surfaces
Governed actions, mobile companions, and private control rooms
Website factory
Live site maintenance driven by signals, review, and weighting
Embodied R&D
3D inspection, movement studies, and physical-digital interfaces
Live surfaces
18+
Build lanes
14+
Public focus
Systems behind systems
Core habit
Automate the repeatable
Public posture
Strong public proof. Tight private boundaries. Real systems behind the glass.
No company secrets on displayNo dangerous control panels exposedNo raw prediction mechanicsNo privileged local paths or private account details
What this is
A project page that acts like a working map.
The common thread is turning messy, high-friction decisions into clear, testable systems, then building the tooling, recovery loops, and operator surfaces that let the next useful system happen faster.
Foundations
Operator fabric, recovery, embodied R&D, and public-safe system posture.
Lineage
How Beat Harbor expanded into media systems, memory, mobile, and site operations.
Projects
The main public products, tools, and vertical systems that carry the work outward.
Runway
Live demos, next unlocks, and the public-safe view of what is actively moving.
Operating thesis
Quiet surfaces, serious internals.
I like public pages that feel composed and legible, even when the systems behind them are doing much heavier work.
Serious internals
The private layer does the hard routing, memory, and automation work.
Operator systems, prediction software, recovery loops, and memory infrastructure stay behind the scenes where they can be governed safely.
Taste under pressure
Design is part of the operating system, not decoration after the fact.
Every surface should help the user understand what is happening, what matters next, and where the boundary between public and private belongs.
Capability mosaic
The work branches, but the design language stays calm.
This is the public-safe snapshot of the core capability families: media systems, story systems, memory systems, weighted automation, embodied interfaces, and device presence.
Realtime media systems
Desktop audio, low-power speech, media ingest, clipping, tagging, and structured recall grew out of one serious realtime engine.
Narrative engines
Longform YouTube stories, steerable interactive books, generated imagery, and context-stable podcast hosts all came from continuity becoming a first-class concern.
Memory and continuity
Short-term and long-term memory systems for Codex turned prior work, decisions, and project state into callable infrastructure instead of lost context.
Weighted automation
The website factory and prediction lanes decide, route, review, and maintain many public surfaces with much less manual intervention.
Embodied interfaces
3D viewers, voxel direction work, and movement studies explore how digital systems can feel inspectable, physical, and controllable.
Device presence
Operator companions, mobile relay lanes, and smaller AI surfaces make the system feel present beyond a single workstation.
Foundations
The systems behind the systems.
The visible apps matter, but they came out of deeper work: operator infrastructure, predictive recovery, embodied 3D, portable AI companions, and tooling that makes the next build safer.
01
Operator fabric
We built one visible operator system across many lanes, machines, and models.
MCP Full Auto and MCP OS grew beyond single-script automation into a control-room idea: one kernel, multiple worker lanes, delegated execution, peer-machine routing, remote chat, and governed actions.
delegated work requestspeer-machine routingremote chat/session storegoverned approval flow
This is the system behind Harper Relay, FlowOS approvals, private command surfaces, and the work that has to keep moving after a public page ends.
02
Recovery intelligence
We pushed automation toward predictive recovery instead of brittle hard stops.
Predictive mesh recovery work focuses on bounded recovery loops that can reinterpret machine state, adjust posture early, and recover before a workflow collapses into a dead end.
predictive mesh recoverymesh brainrouting posture changesverify and reflex lanes
That makes larger automations more durable and gives the bigger systems a better chance of staying useful under real operating friction.
03
Physical-digital bridge
We did real R&D around embodied 3D, movement, and physical inspection.
The 3D work is not just decorative rendering. It includes dense voxel direction work, close-up third-person control studies, layered physical-part inspection, and spatial interfaces that try to make behavior and movement legible.
voxel direction lockthird-person hero framingSanta Fe layered interior inspectionBlender and Three.js build paths
This is the bridge between digital interfaces and physical understanding: how something moves, fits, opens, breaks apart, or gets explored by a person.
04
Device companions
We attached AI and operator context to smaller, more personal surfaces.
ItsMe MicroAI, the Harper Relay operator app, and the ChatGPT connector work all push toward portable context, direct operator presence, and smaller surfaces that stay connected to the real program.
LM Studio-ready micro identitynative-app-friendly bearer authRealtime session armingChatGPT remote MCP connector
That is how the systems become companions instead of staying trapped inside one desktop, one browser tab, or one machine.
05
Systems that build systems
We built the tooling that lets the next tool get made faster and more safely.
Predict Labs, Market R, AssetForge, FlowOS, the site factory, and the domain research surfaces all exist to spot signals, prepare assets, review locally, and promote work only when it is ready.
signal intakeasset decompositionlocal review gatesapproved promotion lanes
This is the meta-layer: the infrastructure that makes the bigger public-facing systems easier to create, test, route, and keep alive.
Tooling stack
The tools we built to make the work possible.
These are not side notes. They are the internal services, research lanes, and operator surfaces that made the public products real.
Signal and opportunity engine
Predict Labs / Market R
Continuous research, scoring, site-operations decisions, and local-first promotion gates that decide what deserves a build pass.
Asset and reconstruction factory
AssetForge
Source-faithful intake, decomposition planning, review metadata, and controlled output for visual work that needs evidence instead of guesswork.
Approval and control-plane surface
FlowOS
A practical review layer for governed requests, operator actions, and infrastructure approvals that should not live only in terminals.
Automation kernel
MCP OS / MCP Full Auto
The runtime substrate for host control, workflow memory, delegation, remote chat, peer-machine routing, and operator visibility.
Embodied simulation and inspection
Santa Fe 3D Viewer + Voxel R&D
High-fidelity inspection tooling, layered interiors, and controllable dense-world studies that explore how physical systems should feel and move.
Portable context surfaces
ItsMe MicroAI + device lanes
Small AI and device-facing layers that carry memory, identity, and operator context onto more personal or mobile surfaces.
Systems atlas
A visual proof of how the work becomes real.
This is the stack in one view: invention and research on the left, enabling systems in the middle, and the public or operator-facing surfaces on the right.
One operator fabric
Many lanes, machines, and models
Automation posture
Recover early, promote carefully
Physical-digital bridge
3D, movement, inspection, and control
Invent
Predictive mesh recovery
Build
MCP OS / MCP Full Auto
Surface
alexharper.us and route lanes
Research and invention pressure-test the idea.
System builders make the idea durable.
Live routes prove it can carry real work.
Invent
01
Predictive mesh recovery
Bounded recovery logic that shifts routing posture before a workflow hard-stops.
Embodied 3D and movement R&D
Voxel direction, third-person framing, layered part inspection, and physical interaction studies.
Portable AI context
Micro identity capsules and operator presence designed for smaller or more personal devices.
Build
02
MCP OS / MCP Full Auto
The operator kernel with workflow memory, governed actions, delegation, and remote chat.
Predict Labs and Market R
Signal intake, scoring, local review gates, and promotion lanes for practical site work.
FlowOS and AssetForge
Approval surfaces and asset factories that make the next system easier to produce safely.
Surface
03
alexharper.us and route lanes
The public narrative layer for portfolio proof, demo-flow, remote-desktop, system-piping, and iamalive.
harperrelay.com and letters.alexharper.us
The customer-facing service surface paired with the operator companion and mobile relay lane.
domain8.me and vertical products
Decision surfaces and productized service lanes where the research and operator work become useful to other people.
Breakout lineage
How Beat Harbor turned into a much larger system family.
This was not one music app with unrelated side quests. It was a seed project that kept breaking open into media intelligence, storytelling, memory infrastructure, mobile companionship, and eventually live operating systems for websites and service businesses.
01
Seed project
Beat Harbor started as a serious realtime audio product, not a toy music sketchpad.
The first push was a desktop-first Rust and TypeScript platform where realtime audio, scheduling, analysis, recording, and creator-first UX all had to coexist without falling apart.
That forced work on low-latency media handling, analysis pipelines, reusable asset libraries, and persistent project context.
02
Audio intelligence branch
That audio foundation broke out into low-power speech and media-ingest systems.
Once the media pipeline was real, it naturally expanded into a realtime text-to-speech direction, voice-preserving synthesis research, and automatic intake flows that can scan media, clip it, tag it, and organize it into callable libraries.
The practical payoff was a reusable engine for voice, media cleanup, clip extraction, and structured audio recall rather than one-off recordings.
03
Story engine branch
Audio tooling then spilled into longform AI storytelling and publishing systems.
The same media, timing, and continuity pressures led to a longform YouTube story generator, then to an interactive book surface that can generate imagery, page-level motion, and steerable narrative branches while the user redirects the story.
Instead of single outputs, the system became a continuity-aware narrative runtime that can carry scenes, imagery, pacing, and branching choices forward.
04
Podcast studio branch
From there came a podcast studio with persistent hosts, variable episode timing, and recallable context.
The project expanded into a podcast lane that can generate full episode series, keep custom hosts stable over time, vary timing and cadence per episode, and call back prior work so the show grows instead of resetting every session.
This made continuity itself into product infrastructure: reusable hosts, accumulated context, and long-running show memory.
05
Memory and runtime branch
That continuity problem fed directly into the Codex short-term and long-term memory system.
To support larger creative and operational chains, I built memory layers for Codex so prior decisions, project context, and working state could stay callable instead of disappearing between runs.
That memory work then supported broader runtime cleanup, including the lag-reduction work that made Codex feel much lighter in heavier desktop and office-style workflows.
06
Device companion branch
Once the memory system worked, the obvious next step was getting Codex onto smaller personal surfaces.
That sparked the iPhone companion direction: bringing the same context-rich operator experience onto mobile, even though the publishing path stayed constrained by Apple developer-account limits.
The important part was proving the operator and memory model could leave the desktop without losing identity or continuity.
07
Operations branch
The same pattern then jumped into service businesses, RoofersChoice, and the website factory.
RoofersChoice and related service systems started as proof-of-concept verticals, then fed into the website factory: a live maintenance and decision system that keeps many sites updated in realtime through weighted automation, review gates, and far less human intervention.
That is where the creative engine became an operating engine: signals, decisions, automation, and maintenance all feeding active websites instead of isolated prototypes.
Project highlights
Public products, private infrastructure, and the bridge between them.
Domain systems
Domain8 and Domain Ledger
Search-first domain research, route governance, and buyer-safe outbound paths that turn raw opportunity signals into public decision surfaces.
Where it excels
Best when messy opportunity research needs to become a clear surface that somebody can actually act on.
domain researchpricing comparisontrust reviewsite routing
Open project
Operator systems
Harper Relay
Managed website repair, AI visibility audits, request routing, and human handoff systems built on top of a deeper operator fabric.
Where it excels
Best when the public site needs a calm front door while the real work continues through private operator lanes.
managed websitesAI visibility auditoperator deskrequest routing
Open project
Side project lab
Extensions, prototypes, local AI, and Living Rooms
A practical R&D bench for Chrome workflows, local models, embodied 3D experiments, micro-device AI, QA loops, and playful spatial prototypes.
Where it excels
Best when a raw idea needs to become something testable fast: an interface, a movement study, a browser workflow, or a small game-like tool.
browser automationlocal model training3D prototypingQA loops
Open projectSide project index
The useful experiments.
These are the things that make the main systems sharper: extension workflows, QA loops, prototypes, private consoles, and model experiments.
Chrome extension
Lead Requote OS
A full insurance lead workflow with cockpit, dashboard, deterministic routing, local model experiments, and repeatable QA passes.
Excels at turning messy lead follow-up into a structured operator flow with testable routing rules.
Identity layer
ItsMe MicroAI
A small personal AI surface with runtime routes, metadata, and a watch/growth loop designed around continuity rather than spectacle.
Excels at making personal context durable enough to reuse across tools and sessions.
Interactive prototype
Living Rooms
A room-building concept with art direction notes, generated references, and a rich creation surface for spatial play.
Excels at turning taste, mood, and constraints into a tactile interface people can explore.
Realtime review tool
Design Director
A browser-facing design review loop with recording endpoints and operator feedback for keeping UI quality from drifting.
Excels at catching visual and UX problems while the project is still cheap to correct.
Private workspace
Command Center
A secure operator surface with voice session access, workflow controls, and work-gallery tracking behind the public company layer.
Excels at keeping private operations powerful without making public pages feel like internal software.
Build system
Site Factory Automation
A queue of review packets, freshness ledgers, route-aware copy work, and repeatable website build passes.
Excels at making repeated website improvements feel governed instead of improvised.
Asset factory
AssetForge
A local-first asset pipeline for source-faithful intake, decomposition planning, review metadata, and approval-aware output.
Excels at turning reference-heavy visual work into a governed job instead of a loose creative scramble.
Automation runtime
MCP OS / MCP Full Auto
A Windows-first operator runtime with host control, workflow memory, governed actions, and a long-term move from MCP Full Auto into MCP OS.
Excels at turning desktop observation, tool routing, and approval-sensitive actions into one readable automation layer.
Desktop music platform
Beat Harbor
A desktop-first Rust and TypeScript music creation platform focused on heavy bass workflows, creator-first composition, and a serious release candidate path.
Excels at turning creative audio ideas into a real desktop product lane instead of leaving them as scattered experiments.
Vertical growth platform
RoofersChoice
A roofing growth stack with estimator flows, branded proposals, payments, financing hooks, CRM workflows, and mobile follow-through.
Excels at translating a single service-business niche into a full operating offer instead of a thin landing page.
Concept to practical use
The part I care about is the handoff from interesting idea to useful thing.
01
Concept
A domain search should feel like a buyer brief, not a registrar funnel.
Take a crowded buying decision and make the first useful answer obvious: first-year cost, renewal math, privacy, and where trust drops.
Shipped form
Built into Domain8 and Domain Ledger as search-first boards with coverage states, evidence-led summaries, and neutral routing.
Practical use
Useful for founders, agencies, and operators who need the real cost picture before committing to a provider.
Open the search flow02
Concept
A website service should hand off like a real desk, not a vague agency promise.
Turn managed website work into visible service lanes with clear trust boundaries, review paths, and honest next actions.
Shipped form
Built into Harper Relay as managed websites, audit intake, request routing, and operator-backed follow-through surfaces.
Practical use
Useful when a business already has a live site but needs clearer offers, steadier intake, and less friction after someone clicks.
Open the managed lane03
Concept
A side project should become testable before it becomes elaborate.
Start with a rough interaction or workflow idea, then make it concrete fast enough to pressure-test whether it deserves a bigger build.
Shipped form
Built into Living Rooms, Lead Requote OS, and the local model bench as tactile prototypes, operator tools, and repeatable QA loops.
Practical use
Useful for turning a loose prototype instinct into something a user can try, critique, and improve the same day.
Open the prototype labMarket R
Prediction software that chooses what is worth building.
Market R is the upstream intelligence engine: public-source capture, prediction scoring, trend ledgers, model ladders, and build queue governance for the website factory.
01
Protocol
Private prediction protocols with public-safe outcomes.
Market R captures public sources, evaluates signals, promotes winning theories, and preserves a private archive so later models can learn from the trail.
02
Scoreboard
Scoreboard-backed work without exposing the playbook.
The private board separates performance by internal criteria, confidence, and readiness while this page only shows sanitized proof of active traction.
03
Build queue
Prediction output feeds Domain8 governance.
Domain8 mirrors Market R queue items into opportunity, site-factory, optimization, and copy-refresh lanes so forecasts turn into reviewed website work.
Concept to practical use
From market signal to shipped website work.
Capture
Public sources and private watchlists
A scored internal signal with timestamped context
Decide
Scoreboard movement and confidence checks
A private action item with a practical lane
Ship
Domain8 route, copy, or optimization target
A reviewed surface improvement visitors can use
Live receipt and sample
A public-safe proof snapshot beside a plain-English example of how a private forecast becomes an action.
Scoreboard receipt
Updated April 23, 2026, 5:19 PM PDT
Lifetime signal
7k+
Active signals
Private
Public-safe opportunity lane
Internal readiness confirmed. Details withheld.
Sanitized action card
A private Market R signal identifies a public page improvement worth testing without revealing the underlying forecast, timing, or source mix.
Confidence
Private
Score
Qualified
Translate the signal into copy, route QA, and a small reviewed optimization pass.
Public boundary
Market R is shown as a portfolio project, while the trade-secret operating details stay off the public page.
No exact call cadence
No source weighting
No raw scores
No trigger mechanics
Now shipping
Ongoing work, not just archive pieces.
The portfolio is meant to show what is live, what is being tightened, and where the next useful pass is happening.
01
Active build
Alex Harper portfolio surface
Reworking the homepage into a true systems breakdown: not only what shipped, but the enabling tools, operator fabric, and R&D that made the visible products possible.
Updated April 23, 2026
02
Prediction engine
Market R opportunity and prediction software
Private source capture, model scoring, trend ledgers, prediction scoreboards, and mirrored build-queue governance feed the website factory.
Scoreboard updated April 23, 2026
03
Operator refinement
Harper Relay trust and intake lanes
Continuing to sharpen service handoff, AI visibility entry points, and the boundary between public clarity and private operations.
Tracked in current route work
04
Prototype lab
Living Rooms, extension tooling, and local model loops
Keeping the side-project bench active so rough ideas become testable interfaces, movement studies, workflows, and small systems quickly.
Ongoing experimental lane
05
Integration target
ChatGPT account memory and project history
The ChatGPT side should become a first-class source for ideas, decisions, project notes, and reusable context once an export or connector is available.
Same account, export not connected yet
06
Operator platform
MCP OS and MCP Full Auto consolidation
Evolving the Windows-first desktop operator runtime, governed action system, host kernel, and operator console from MCP Full Auto into MCP OS.
Tracked in OneDrive runtime workspace
07
Product lane
Beat Harbor desktop music platform
Pushing a creator-first desktop audio product with Tauri, TypeScript UX, import and remix workflows, and a release-candidate mindset.
Desktop-first v1 path active
Live surfaces
The domains that already carry the work.
These are the public surfaces already doing real portfolio work: hosting the story, carrying the offer, or turning research into a usable interface.
01
alexharper.us
Alex Harper live portfolio
The public map of systems, tools, experiments, and ongoing work, now used as the main proof surface instead of a static about page.
Acts as the public index where local programs, hosted lanes, and future integrations get promoted into one narrative.
02
harperrelay.com
Harper Relay business OS
A live operating layer for lead management, request routing, context carryover, and human-aware scheduling across service workflows.
Shows the customer-facing and operator-facing promise of Harper Relay in one domain instead of splitting the story across private notes.
03
letters.alexharper.us
Operator companion and mobile relay lane
The live operator-app hosting lane for direct MCP-backed chat, state polling, and the phone-side companion surface to the larger operator system.
Shows that the private runtime already has a smaller, more personal access point instead of staying trapped in desktop-only infrastructure.
04
domain8.me
Domain8 research and search surface
A search-first domain buying surface with pricing context, renewal math, trust review, and operator-ready research framing.
Makes domain research feel like a decision product rather than another registrar funnel.
Route map
Named surfaces inside the main domain.
The domain is not only one homepage. It already contains route-level proof of the operator, simulation, and systems-wiring ideas behind the broader portfolio.
alexharper.us/demo-flow
Concept handoff lane
A public-facing route for showing how an idea moves toward a usable flow instead of staying trapped in planning notes.
alexharper.us/remote-desktop
Remote control and device-presence lane
A route family that points toward desktop presence, remote operation, and machine access as product surfaces instead of hidden admin tricks.
alexharper.us/system-piping
Systems wiring lane
A route that frames the connective tissue itself as part of the work: how systems hand off, route, and stay legible.
alexharper.us/iamalive
Voxel and world-simulation lane
A live route tied to the dense-world, movement, and controllable-environment experiments that support the embodied side of the portfolio.
Demo runway
The next demos already have jobs.
3/4 active
Updated April 24, 2026
4 demo lanes are on the board: 3already active and 1 still waiting on the next approval or integration step.
Active
3
Planned
1
Status
75% ready
3 active lanes1 planned lane
Active covers public-safe live work, local operator demos, and private working tools. Planned stays reserved for integrations still waiting on approval or a connector.
Next unlock
ChatGPT project memory bridge
Connect an export or approved account source, then show only sanitized project summaries.
Public-safe live
1
Local operator demo
1
Private working tool
1
Planned integration
1
Runway board
A visual map of what is already moving, what stays internal, and what still needs one more unlock before it becomes stronger public proof.
Public view
1
Local view
1
Private view
1
Future view
1
Lane 01
Market R to public-safe opportunity cards
Public-safe live
Sanitized receipt
Public view
Market R section on this page
Readiness 4 of 4
Signal
Sanitize
Ship
Lane 02
FlowOS as the approval surface
Local operator demo
Local admin route
Local view
FlowOS /admin/control-plane
Readiness 3 of 4
MCP request
FlowOS review
Governed action
Lane 03
ChatGPT project memory bridge
Next unlock
Planned integration
Account-context lane
Future view
approved export-backed project memory
Readiness 1 of 4
Conversation
Approved export
Project memory
Immediate move
Connect an export or approved account source, then show only sanitized project summaries.
Lane 04
Lead Requote OS from concept to operator tool
Private working tool
Extension package
Private view
extension cockpit walkthrough
Readiness 3 of 4
Lead context
Cockpit route
QA loop
Readiness 4 of 4
Live demo candidate
Market R to public-safe opportunity cards
Public-safe live
Sanitized receipt
01
Signal
02
Sanitize
03
Ship
Concept
Prediction work is only useful if it turns into a better decision without leaking the private playbook.
Working demo
The portfolio now shows sanitized receipts, withheld mechanics, and a clear path from signal to website work.
Practical use
Lets visitors understand the business value of prediction software while keeping cadence, weighting, raw scores, and triggers private.
Next: Promote one more sanitized before/after build result once it is safe to publish.
Open next: Public view: Market R section on this page
Readiness 3 of 4
Live demo candidate
FlowOS as the approval surface
Local operator demo
Local admin route
01
MCP request
02
FlowOS review
03
Governed action
Concept
Governed actions should not be hidden in a terminal when the operator already lives in a workflow app.
Working demo
FlowOS now has an admin Control Plane route that can read pending MCP host requests through a local app proxy.
Practical use
Makes domain verification, infrastructure approvals, and future governed operations easier to review from one place.
Next: Add a public-safe screenshot or diagram once the approval executor path is fully confirmed.
Open next: Local view: FlowOS /admin/control-plane
Readiness 1 of 4
Live demo candidate
ChatGPT project memory bridge
Planned integration
Account-context lane
01
Conversation
02
Approved export
03
Project memory
Concept
Useful decisions from conversations should become project memory instead of disappearing into search history.
Working demo
The current portfolio names ChatGPT integration as an account-context lane and leaves room for export-backed receipts.
Practical use
Turns long-running planning threads into reusable evidence for portfolio updates, decisions, and next-step queues.
Next: Connect an export or approved account source, then show only sanitized project summaries.
Open next: Future view: approved export-backed project memory
Readiness 3 of 4
Live demo candidate
Lead Requote OS from concept to operator tool
Private working tool
Extension package
01
Lead context
02
Cockpit route
03
QA loop
Concept
Insurance lead follow-up needs a cockpit, repeatable routing, and QA feedback, not another loose spreadsheet.
Working demo
The extension package includes cockpit, dashboard, deterministic routing, local model training, and QA notes.
Practical use
Helps turn lead chaos into a testable browser workflow with safer review loops before bigger automation.
Next: Add a short visual walkthrough that highlights the cockpit without exposing private lead data.
Open next: Private view: extension cockpit walkthrough
Trust and hardening
Live work should show its guardrails.
A portfolio for active systems needs proof that the demos, queues, and project notes are being handled with practical security habits.
Exposure check
Public page copy avoids local paths and private account handles.
The portfolio can mention active work without exposing workstation directories, raw account identifiers, or export locations to visitors.
Route boundary
Privileged build-queue data stays loopback-only in development.
Local operator routes can still be tested from the machine, while non-loopback requests are refused before private queue details leak.
Browser posture
Security headers are part of the live surface now.
The page ships with a stricter CSP, framing controls, content-type protection, referrer limits, and permission denials for unused device APIs.
Developer workspace
The active bench behind the portfolio.
These are ongoing projects in the local Developer workspace. They broaden the portfolio beyond this site into account context, asset tooling, audio, copy systems, search surfaces, and local AI context.
Continuous optimization stream
Predict Labs
OneDrive Documents / MCP Full Auto / predict-labs
The canonical home for prediction-driven growth, site-operations work, asset lookup, local publish review, and specialist-bot handoff discipline.
Turns loose research threads into one continuous stream that senses demand, builds locally, checks safety, and routes work to the right live destination.
Prediction software
Market R
Local workspace / market-r
A TypeScript and Python prediction workspace with ingest, watchdog, scoring, market workers, runtime DB scripts, a control plane, trend ledgers, and prediction artifacts.
Turns public market signals into scored opportunities that can trigger build queues, site-factory work, and practical service products.
Runtime cognition and recovery
Predictive State Mesh Lab
OneDrive Documents / MCP Full Auto / predict-labs
A focused lane for predicting operator and runtime state inside MCP OS, separate from market work and aimed at desktop cognition, routing, and recovery.
Keeps the operator runtime adaptive instead of reactive by treating state drift as something that can be modeled and corrected early.
Account context
ChatGPT Integration
Same OpenAI account / export pending
A planned bridge from ChatGPT conversations into the same live portfolio map, so concepts, decisions, and recurring project threads can show up beside local code work.
Turns useful conversation history into evidence-backed project memory instead of leaving it trapped in chat search.
Mobile operator companion
Harper Relay Operator App
OneDrive Documents / MCP Full Auto / mobile / harper-relay-operator
A direct mobile surface for MCP Full Auto with login, state polling, chat selection, message send, and backend realtime-session arming.
Brings the operator system onto a phone and proves the runtime can survive outside the desktop shell.
Approval and operating surface
FlowOS
Local workspace / FlowOS
A workflow app being extended into the place where governed MCP requests, approvals, and operator control-plane actions can be reviewed.
Keeps infrastructure decisions, account actions, and recurring system approvals inside a practical app surface instead of scattered terminals.
Physical inspection tooling
Santa Fe 3D Viewer
OneDrive Documents / MCP Full Auto / projects / santafe-3d-viewer
A high-fidelity iPhone-first 3D inspection kit with layered internals, per-part explanations, alternative parts, exports, and optional Blender automation.
Turns physical systems into navigable digital inspection surfaces that make materials, movement, and repair context easier to understand.
Asset pipeline
AssetForge
Local workspace / AssetForge
A local-first asset factory for source-faithful intake, reconstruction planning, decomposition, review metadata, and QA gates.
Turns a reference image into a governed job with visible locks, inference zones, artifacts, and review state.
Embodied world and movement R&D
iamalive / Voxel Dominion
OneDrive Documents / MCP Full Auto / docs and game direction
Dense voxel terrain direction, third-person hero-follow framing, and physical world-sculpting verbs used to study scale, movement, and controllable environments.
Pushes the physical-digital frontier by testing how a person, organism, or operator should feel moving through a dense controllable world.
Search product
Marketplace Search Console
Local workspace / MarketplaceSearchConsole
A Next 16 workspace for marketplace search and console-style decision surfaces.
Useful when marketplace data needs to become a scannable operator surface rather than another generic results page.
Desktop audio tool
WubForgeV2
Local workspace / WubForgeV2
A Tauri and desktop workspace for music production tooling, milestone QA, and voice-lab experiments.
Turns creative audio workflow ideas into desktop app surfaces with build, typecheck, and milestone verification paths.
Service copy system
Copywriter
Local workspace / Copywriter
A service-business copy library with scripts, optimization guides, and bot briefs for trades and lead-generation workflows.
Gives local service businesses practical sales, rescue, follow-up, and conversion copy that can improve with real notes.
Local identity capsule
ItsMe MicroAI
Local workspace / ItsMeMicroAI
A small inspectable context layer that grows from local project evidence and exports an LM Studio-ready system prompt.
Keeps working patterns, taste, and project context portable without mixing in unrelated model-training data.
Desktop operator runtime
MCP OS / MCP Full Auto
OneDrive Documents / MCP Full Auto
A Windows-first autonomous operator stack with desktop observation, governed action routing, workflow memory, and a Rust-first host-kernel direction.
Turns desktop automation into a real product and infrastructure lane instead of a pile of one-off scripts.
Desktop music creation platform
Beat Harbor
Local workspace / WubForgeV2
A desktop-first Rust and TypeScript music creation platform oriented around bass-heavy creation, creator-first UX, and a serious release candidate.
Turns audio tooling ideas into a focused product lane with real execution boundaries and milestone discipline.
Vertical service platform
RoofersChoice
Local workspace / RoofQuote
A roofing growth platform with customer estimator flows, branded proposals, financing hooks, internal CRM workflows, and mobile support plans.
Shows how one vertical can become a deep, productized operating system instead of a generic lead form.
Windows installer lane
AI Buying Assistant Package
Local workspace / ai-buying-assistant-package
A packaged Windows install and verification artifact for an AI buying assistant app, built as a prep layer for installer hardening and distribution.
Turns internal app work into something that can be installed, verified, and shipped beyond a dev machine.
GitHub one-offs
Quick repos become portfolio proof.
This lane is for repo-backed experiments that should graduate into useful demos, not just sit as forgotten one-offs.
01
Verified remote
domain8
Connected GitHub remote / domain8
The main live web surface and portfolio host, with hostname routing, project pages, generated visuals, and ongoing public-site experiments.
Acts as the place where strong one-off ideas get promoted into public, testable portfolio surfaces.
02
Local repo placeholder
FlowOS
Local workspace / FlowOS
A local Git workspace with remote wiring still using a placeholder URL, useful as an early-stage systems concept lane.
A candidate for turning workflow and operating-system ideas into a clearer demo or repo-backed project story.
03
Account source
GitHub one-offs
Connected GitHub account
Small repositories, experiments, and one-off builds that should be harvested into this portfolio as evidence-backed demos instead of staying scattered.
Keeps quick builds from disappearing by giving each useful repo a lane: concept, shipped form, practical use, and next step.
Working style
Taste plus execution, with receipts.
- 01Find the real user decision.
- 02Build the smallest complete surface.
- 03Instrument the handoff and trust boundary.
- 04Test with screenshots, routes, and behavior.
- 05Keep the thing alive after the first ship.