Portfolio City

Bismarck AI is an experiment in agent orchestration and harness UI/UX presented as a Victoria II/EU5-style grand strategy game, set in a distinctly fictional world of anthropomorphic walruses fighting swarming bugs controlled by the Froggy Empire.

LLM cost intelligence with a maintained API cost calculator, calcLLM: Trace for production LLM/API costs, and calcLLM: Crew for AI coding-tool costs. The calculator estimates frontier-model token spend, history, reasoning, caching and long-context pricing. Trace remains planned; Crew is synthetic-tested and awaiting live provider qualification.

AI form-builder code and research preserved after the product was killed before deployment: crowded commodity economics and weak differentiation did not justify the compliance, integration, infrastructure, and customer-support burden.

An always-on workstation for AI coding agents, with persistent sessions and an Agent Board for monitoring and control. The MVP is implemented; VPS reliability, durable history, and Mac-to-phone continuity are undergoing validation. Our first brand rebirth: gBlockParty now names the agent workstation; its original content platform is archived.

OSS desktop front-end for local and GBlockParty-managed agent/browser workspaces.

Clustered with gblockparty

GSkillPacks' current commercial product thesis was killed after its alignment- and interrogation-heavy workflow imposed too much ceremony. The public agentic-skills repository, published skillpacks package, domain, and reusable workflow library remain preserved as active OSS/shared infrastructure; an alignment/interrogation-less v2 has not yet been activated.

AI coding subscription pacing app for Claude, Codex, Gemini, and future providers; Pitwall Local is in technical GitHub beta/source-install readiness, while the monorepo preserves the path to a future CalcLLM-powered connected edition with opt-in cloud sync and team cost dashboards.

Owned domain awaiting mission assignment. No codebase, no users — held in reserve.

Code snippet manager prototype preserved after the standalone commercial thesis was killed: a real Clerk, tRPC, semantic-search, Gist-import, billing, and VS Code extension foundation never established validated demand or enough differentiation to justify launch and ongoing support.

Config-as-code PostHog implementation and research preserved after the standalone product thesis was killed: computer-use automation absorbed most click-ops, no TrackCTL-specific demand was validated, and the remaining cross-cutting secrets problem belongs in the EnvBank experiment.

AI-powered raster-to-vector conversion for Glowforge, xTool, and Cricut fabricators; dormant because it is not currently testable and should not be treated as live revenue work.

Validation-stage resilient browser-session control plane for AI agents — one versioned JSON contract planned across CLI, HTTP API, and MCP adapters, with semantic snapshots, explicit reconnect/resume behavior, and causal replay

AI coding-tool cost intelligence for engineering leaders, with per-person and team breakdowns, budgets, and reporting. The application is implemented and tested with synthetic data. Codex, Cursor, and Claude Code adapters await live qualification; the pilot remains blocked on eligible administrator access and qualifying usage.

Clustered with calc-llm

Production LLM and API cost intelligence for startup CTOs — answers why the bill is 3x what token pricing implies, within the calcLLM product line

Clustered with calc-llm

Block-based web terminal library — five-package TypeScript/React monorepo (core, parser, block-model, transport, react) with CI

Auto-generated, auto-maintained product tutorials and videos from a GitHub repo plus staging app — YAML-defined flows regenerated in CI/CD for DevRel teams

Mobile remote control for AI coding agents over SSH — Expo/React Native client with a tmux session UI for monitoring and driving Claude Code sessions from a phone

AI-powered SVG animation tool — describe an animation in text and get SMIL/CSS output, backed by Gemini

Visual review and monitoring for agent-proposed Terraform deployments. Inspect infrastructure changes and diagnostics in interactive AWS architecture diagrams, review saved plans, and follow Terraform apply progress. Agent plugin packaging and approval-workflow integration are planned.

TypeScript SDK for multi-agent AI apps — DAG task decomposition, parallel agent execution, token budgets, and streaming

NixOS-based operating system optimized for agentic coding — CLI agents, headless agent desktop, CDP browser control, and computer use from one flake building a QEMU VM, ISO, or WSL image.

Podcast and video platform for Boston founders — SDK and schema foundation shipped; archived 2026-04-27 before Stripe membership billing was implemented.

AI changelog platform with GitHub/GitLab webhooks, release management, email digests, monorepo support, and publishing workflow — release-card extraction and list API cleanup remain before beta

Deferred post-v1 Spellbook OS planning module for brands, campaigns, shows, episodes, objectives, CTAs, and channel-intent capture

Clustered with spellbook-os

Research-first content-repurposing brand for solo creator-operators. v1 is Prism only: ingest one long-form source, extract moments, generate channel-native derivatives, and review outputs. Foresight, Portals, Ritual, Sanctum, and Scry are deferred post-v1.

Deferred post-v1 Spellbook OS channels module — destination accounts, provider setup, token health, and channel capability surfaces

Clustered with spellbook-os

Sole current Spellbook OS v1 focus — ingest one long-form source, extract moments and hooks, generate channel-native derivatives, and review outputs before export

Clustered with spellbook-os

Deferred post-v1 Spellbook OS publishing module — approval queues, scheduling, publishing, retry, and derivative-asset export

Clustered with spellbook-os

Deferred post-v1 Spellbook OS workspace hub — health, next actions, queue pressure, connector status, and top signals across the long-term module suite

Clustered with spellbook-os

Deferred post-v1 Spellbook OS analytics module — performance, patterns, recommendations, and feedback for future planning cycles

Clustered with spellbook-os

Cyberpunk-themed OBS stream overlay showing live Stripe revenue and cost bars for the build-in-public journey — Bun server with themes, admin panel, and Stripe org key support

Structured script editor for YouTube creators — parse scripts into blocks, edit, splice, run a teleprompter, and keep immutable revisions on a Next.js/tRPC/Drizzle stack

Screen-recording tool in the CamStudio lineage for content capture.

Local, no-cloud pipeline to transcribe, OCR-narrate, and build a reverse script from screen recordings.

Canonical narration-first production dashboard and recording console for George's three technical Shorts series, with episode status, packet provenance, narration review, and teleprompter workflows.

Boston startup community platform that shipped its planned scope, then died after launch momentum faded, founder/investor pull weakened, event attendance declined, and the required outreach/follow-up became the wrong channel for the founder.

Government-data SaaS for municipal efficiency scoring; dormant and not functional until municipal financial data collection and the RAG pipeline exist.

Mobile-first fantasy stock draft game validating a fixture-backed path from stock search through a first pick and board preview; standings, sharing, accounts, live market data, and a packaged application remain deferred

A practical experiment in connecting third-party services for personal administration. Built around subscription and renewal workflows, JIT Life OS explores how APIs—and potentially agent computer use—can help turn scattered information into useful next steps. A runnable, fixture-backed prototype exists; live integrations remain under development.

Visual firearm builder and community that was killed after payment-provider restrictions made reliable firearms-adjacent monetization economically unviable.

Customer-discovery operating system for technical B2B SaaS founders with private workspace, assumptions, experiments, evidence, claim review, public proof, reporting, and pricing-readiness hooks implemented; ready for a bounded launch as a portfolio asset, with limited expected standalone SaaS upside

Parametric architectural modeling studio for exploded axonometric diagrams — turbo monorepo with a working dashboard prototype (grid snapping, panels, camera persistence) under a freemium SaaS spec

Browser-extension product in discovery — ticketing and customer-service workflows with a creator-monetization angle, shaped through research, specs, and an assumption tracker

Preserved sim-racing coaching research from an ended product thesis: no application shipped or recorded adoption emerged, and the owner concluded that credible racing instruction required professional racing or coaching authority the proposed student-built-for-students positioning did not provide.

Browser-first 3D real-time strategy game (Divisions 1945) — post-V-E-Day order-of-battle research, division/deck rosters, and a shared unit-data schema for the tactical layer, deck builder, and campaign map.

Playable browser football arcade prototype preserved after its standalone commercial thesis was killed. The completed game and deployed build served primarily as a Fable 5 capability experiment, with no validated demand, adoption, monetization, or go-to-market evidence supporting continued product investment.

Substantial RollerCoaster-Tycoon-style terminal simulator preserved after its standalone commercial thesis was killed. The completed Rust prototype and its unlisted 3D browser-port companion served primarily as Fable 5 capability experiments, without release, distribution, adoption, revenue, or go-to-market evidence supporting continued product investment.

Research-backed concept for a mobile-first ranked cognitive arena; no application code exists, and task depth, anti-cheat, ladder liquidity, and willingness to pay remain unvalidated.

Omni Editor web MVP and SwiftUI mobile package skeleton.

A twenty-wave naval fire-control campaign with persistent crew management.

A growing family of top-down arcade war games built around battlefield command, supporting fire, and the consequences of getting too close. Includes Danger Close: WW1 and Danger Close: Thunderbolt.

Western Front artillery game for browser, iPhone, and iPad. Defend your trench, upgrade your battery between assaults, and avoid hitting your own troops. The native iOS/iPadOS version offers a three-wave free campaign and an optional $4.99 one-time Unlimited Campaign unlock; App Store release remains pending.

Clustered with danger-close

Top-down A-10 close-air-support game in a fictionalized 1991 Gulf War scenario. Designate gun runs, Maverick targets, and bomb strikes to stop armored waves while protecting coalition positions. A browser prototype is available; the separate free native iOS/iPadOS build is archived but not recorded as submitted to the App Store.

Clustered with danger-close

Feature-complete browser game engine held in Purgatory while a bounded re-engagement test determines whether acquisition can support revival

Poketo’s CRM teammate for managing contacts, deals, and sales pipelines. Implemented in the v3 agent-first business suite, with shared capabilities for people and AI agents; in development.

Clustered with poketo

Poketo’s planned growth teammate for outreach, campaigns, and analytics. Historical v2 implementation is archived; Growth awaits a rebuild for the current v3 suite.

Clustered with poketo

Poketo video-clipping sub-brand intended to turn timestamp lists and local source videos into landscape and vertical MP4 clips; its dedicated repository remains a pre-build concept scaffold.

Clustered with poketo

Build your business team with Poketo. A modular business suite with friendly animal mascots, designed for people and AI agents to work together across projects, customers, communication, and reporting. In development, with Work, CRM, Report, and Chat/Brain implemented. Poketo owns business records, durable employee memory, and human approvals; EXPLe Corp HQ provides agent execution.

Professional-grade orbital mechanics simulation with an interactive tutorial foundation, now dormant and held for possible future revival.

Browser PWA for playing the board game Go — offline-capable vanilla TypeScript app with puzzles, service worker, and Playwright coverage

Browser turn-based grand strategy game blending Civilization, Victoria 2, and Heroes of Might and Magic

Slack-native burnout detection — dashboard, check-ins, alerts, Slack, AI digests, billing, and email surfaces exist; DB migration, onboarding, webhook verification, and landing page remain before beta

Lightweight IT asset management for growing teams. Track ownership, locations, custom asset fields, and change history, with CSV import/export and role-based access. Core product, billing, onboarding, and notification flows are implemented; production setup and live-service validation remain before self-serve launch.

Tracks agent-built products through the AFPS skills pipeline — experience-map and approval-gate state for products moving through discovery and design.

SaaS platform (in design workflow).

Anno-style isometric city-builder experiment with PixiJS — a game concept exploring mechanics for Bismarck AI (LexDev).

Public war room at leexperimental.com — portfolio dashboard, operations tracker, financial scoreboard, intel, graveyard, and billing ledger for the EXPLe Corp portfolio

EXPLe Corp portfolio launcher — phone-style app grid with boot splash and app-store-style detail drawers linking out to portfolio products