Projects.

Jump Trading

The Playbit ICT Suite covers Fair Value Gaps, Order Blocks, Breaker Blocks, the Inverted Liquidity Model, and full entry signal logic. The NQ/ES ICT Suite v3 layers in PDH/PDL/PWH/PWL levels, Kill Zone overlays, CISD detection, and a session dashboard.

The flagship indicator is the Unicorn Entry Model, designed to identify high reward-to-risk setups on equity index futures. Notably called a 893-point move on NQ at 66:1 reward-to-risk during the Trump ceasefire candle.

Alpha Journal

Most trade journals require manual entry, which is why most traders abandon them. Alpha Journal closes the loop with a three-stage pipeline: Gmail ingestion of broker confirmations, AI parsing into structured JSON, and an authenticated POST to the journal API.

Two delivery modes ship by default. The first is an on-demand Claude Code slash command for traders who want manual control. The second is a fully autonomous Make.com scenario that runs every five minutes, keeping the journal current without intervention.

The Apothecary Brain

166 remedies, 148 herbs, and 357 cross-linked notes, all built from 25 source PDFs into a fully navigable Notion knowledge base. Four deployed AI agents handle Personal Apothecary consultations, Morning Wellness Routines, What's In My Kitchen remedies, and Seasonal Apothecary planning.

The build process became its own product: the PDF-to-Brain Pipeline, a repeatable system for converting any knowledge corpus into a productized AI-augmented Notion base. The Apothecary is the proof of concept.

DarkLedger

Fourteen pages of operations tooling, including a Strategy Map built on React Flow that allows traders to visualize the relationships between setups, instruments, and outcomes. The interface is dark by design, optimized for long sessions in front of charts.

Designed as a freemium commercial product, distinct from Alpha Journal, with a focus on professional-grade trading operations rather than journaling alone. Public release planned.

JumpX Studio

The core workflow is VoxelPlot storyboard-to-video. Sequential reference image generation — where each frame anchors the visual identity of the next — solves the consistency problem that single-prompt video generation cannot. Output runs through Higgsfield and Seedance 2.0.

Documented constraints: Seedance handles solo subjects and atmospheric environments cleanly but breaks on multi-vehicle physics. Brand and military terms reliably trigger content filters and require generic descriptors as workarounds.

Dispatch

Investigations into the trades, structures, and ownership games that fall between mainstream finance coverage and crypto-style hot takes. Each volume documents a specific market event with primary-source data, technical breakdowns, and a focus on the structural mechanics most coverage misses.

Volume 01 covers the Avis Budget Group squeeze and the cash-settled swap loophole that made it legal.

3D Interactive Rescue Site

Visitors orbit a stylized village rendered in real-time WebGL, encounter five animated characters with full walk cycles, and tap any of five interactive locations to open data-rich modals with calls to action. Custom GLSL shaders animate 22,500 individually wind-blown grass blades. Raycasting handles tap detection. Dynamic lighting and shadows render at 60 frames per second in the browser.

The entire experience ships as a single 120KB HTML file with zero build step. Production stack: Three.js r128 loaded via CDN, Meshy AI for every 3D asset, gltf-transform CLI for 40-60x asset compression, Claude Code for the code itself, and Replit for static hosting. Approximately 2,300 lines of code, produced through end-to-end AI collaboration in a single overnight build session.

The build process was documented in full and published as a standalone writeup, including every Meshy prompt, every shell command, and every Claude instruction used. The same pattern works for portfolio sites, product landing pages, interactive explainers, and campaign launch sites.

ASTRO QUANT

Generates daily trade bias signals — Long, Neutral, or Risk Off — across ten assets in four categories: Crypto, Stocks, Indices, and Metals. The unified score fuses lunar phase, planetary retrograde states, derivatives positioning, liquidity flows, and price structure relative to long-term moving averages. Designed as a research and decision-support tool, not an execution platform.

The signal engine is centralized in two TypeScript files. astroLogic.ts handles the deterministic decision layer: lunar phase calculation, eclipse window detection, retrograde tracking, and an aggregate astroSentiment score from −1.0 to +1.0. api.ts orchestrates live data fetching from CoinGecko, Finnhub, FMP, DeFiLlama, and Binance Futures, with full fallback chains for each asset class.

A deliberate research-based inversion drives the lunar component: Full Moon as bullish (peak euphoria), New Moon as bearish (caution and selling pressure). This is layered with a 200 EMA fair value framework that classifies each asset into zones from Deep Discount through Stretched, then weighted by retrograde penalties and liquidity conditions to produce the final bias.

The architecture is intentionally serverless. All computation runs client-side on the user's device. AsyncStorage handles local persistence for journal entries, portfolio data, and watchlists. No backend, no database. Distributed through TestFlight and Google Play via EAS Build, with a freemium model gated through RevenueCat.

PDF-to-Brain Pipeline

Source documents land in an Obsidian vault, are processed and cross-linked locally, then pushed to a Notion workspace via the official API. The output is a navigable knowledge product with attached AI agents that can answer questions in the voice and structure of the source material.

Originally built for The Apothecary Brain. Now operates as a standalone offering for any knowledge niche where a corpus of PDFs needs to become a structured, queryable, monetizable product.

Trading Content System

A dedicated Claude Code workspace anchored by a CLAUDE.md brand brain file, an X algorithm playbook of 651 lines, and a COMMANDS.md registry of eleven custom slash commands. Image generation runs through a custom shell script that wraps the OpenAI API.

A Data Accuracy Rules section was added after early generations produced fictional price levels. The system now enforces source-anchored claims and refuses to invent technical levels not present in the underlying chart data.

Gmail-to-Journal Automation

The pipeline reads broker confirmation emails from Gmail, parses the contents through an AI layer into a normalized JSON schema, and posts the result to the Alpha Journal API. Two delivery modes are supported: an interactive Claude Code slash command and a fully autonomous Make.com scenario that runs every five minutes.

The same pipeline architecture has been deployed for an external trader running a cloud VPS, demonstrating that the system generalizes across brokers and infrastructure environments.

Karpathy-Style Second Brain

Viral content is captured by XPOZ scanners and dropped into a raw intake folder. Claude Code compiles raw material into a structured wiki. The TRIBE v2 engagement scoring system queries the wiki when generating posts, and post performance data feeds back into the wiki to inform the next cycle.

The architecture is intentionally inspired by Andrej Karpathy's notes on self-improving knowledge systems: separate raw, processed, and outbound layers, with feedback closing the loop.

Pool Operations Automation

Currently in active development. The system is being designed to route scheduling, monitoring, and customer communication through a unified agent-driven backend, replacing the manual coordination that dominates the industry today.

Public details to follow as the build progresses.

Notion Agent Stack

Each major venture has a dedicated Notion agent configured with the relevant database schemas, brand voice, and operational rules. Agents handle vertical-specific tasks — content drafting for one brand, customer queries for another, knowledge retrieval for a third — without context bleeding across the system.

X Article Playbook

The playbook prescribes deep multi-search research, a contrarian-hook opening, scannable header structure, short declarative sentences, a screenshot-worthy closing line, and a sources list. Distribution best practices — chart visualization, thread format option, peak US/EU timing windows, account tagging, CTA framing — are encoded as part of the workflow.

The Avis Budget breakdown that anchors the Dispatch series was produced using this playbook end to end.