Software we build, own, and run

 _  _ __   __ ___  ___  ___  ___     _____  ___    _    ___   ___  _  _   ___    ___  _  _   ___  ___  _  _  ___
| || |\ \ / /| _ )| _ \|_ _||   \   |_   _|| _ \  /_\  |   \ |_ _|| \| | / __|  | __|| \| | / __||_ _|| \| || __|
| __ | \ V / | _ \|   / | | | |) |    | |  |   / / _ \ | |) | | | | .` || (_ |  | _| | .` || (_ | | | | .` || _|
|_||_|  |_|  |___/|_|_\|___||___/     |_|  |_|_\/_/ \_\|___/ |___||_|\_| \___|  |___||_|\_| \___||___||_|\_||___|

Overview

A neural network that combines quantitative and qualitative analysis to generate profitable trading signals. Unlike pure quant approaches, this system captures the "soft" factors that numbers alone miss.

Quantitative Inputs

• Financial ratios: P/E, ROE, EPS, revenue growth
• Economic indicators and market data
• Historical price and volume patterns
• Statistical models and backtested signals

Qualitative Inputs (Quantified)

• NLP sentiment scores from news and social media
• Management quality metrics derived from earnings calls
• Brand strength indicators
• Industry trend analysis
• Competitive advantage (moat) scoring

Approach

The system quantifies traditionally subjective factors, turning CEO interview tone, brand perception, and market sentiment into numerical features. These are combined with traditional financial metrics to train a neural network that can identify opportunities pure quant systems miss.

Status

Initial pipeline using Grok API performed well, successfully extracting sentiment signals from X posts and live news sources. Currently developing a new NLP pipeline to expand coverage and improve signal quality.

 _____  ___   ___  _  _  _  _  ___   ___    _    _         _    _  _    _    _    __   __ ___  ___  ___    ___  _  _   ___  ___  _  _  ___
|_   _|| __| / __|| || || \| ||_ _| / __|  /_\  | |       /_\  | \| |  /_\  | |   \ \ / // __||_ _|/ __|  | __|| \| | / __||_ _|| \| || __|
  | |  | _| | (__ | __ || .` | | | | (__  / _ \ | |__    / _ \ | .` | / _ \ | |__  \ V / \__ \ | | \__ \  | _| | .` || (_ | | | | .` || _|
  |_|  |___| \___||_||_||_|\_||___| \___|/_/ \_\|____|  /_/ \_\|_|\_|/_/ \_\|____|  |_|  |___/|___||___/  |___||_|\_| \___||___||_|\_||___|

Overview

A neural network trained exclusively on price and volume data for market timing. Pure technical analysis approach: objective, rule-based, and fully backtestable.

Data Sources

• Historical price data (OHLCV)
• Trading volume patterns
• Price-derived indicators
• Chart patterns and formations

Technical Indicators

• Momentum: RSI, MACD, Stochastic
• Trend: Moving averages (SMA, EMA), ADX
• Volatility: Bollinger Bands, ATR
• Volume: OBV, Volume Profile
• Pattern recognition: Head & shoulders, triangles, trends

Approach

The system learns to identify market psychology and timing purely from price action. It ignores fundamentals entirely, focusing on what the market is actually doing rather than what it "should" do. This makes it objective and scalable.

Backtest Results

The AI strategy significantly outperforms SPY (S&P 500 ETF) buy-and-hold over the test period. SPY serves as the standard market benchmark, so consistently beating it suggests the model is generating alpha (excess returns beyond passive investing).

Monte Carlo simulations show the AI strategy outperforms approximately 99.7% of random trading outcomes. This suggests the results aren't attributable to luck alone and may reflect genuine predictive edge.

Risk Considerations

While returns are promising, the strategy exhibits higher volatility than SPY buy-and-hold, with sharper drawdowns and faster gains. This raises important questions about risk management and position sizing. A complete evaluation would require risk-adjusted metrics like Sharpe ratio (return per unit of risk).

Additionally, real-world factors such as transaction costs, slippage, taxes, and liquidity constraints are not fully captured in backtesting. Whether these results hold out-of-sample remains to be validated through live testing. Past performance is not indicative of future results.

Status

Backtesting complete with promising results. Currently developing the automated execution system for live trading validation.

   _    ___  ___  ___    _    _       _       _    ___  ___
  /_\  | _ \| _ \| __|  /_\  | |     | |     /_\  | _ )/ __|
 / _ \ |  _/|  _/| _|  / _ \ | |__   | |__  / _ \ | _ \\__ \
/_/ \_\|_|  |_|  |___|/_/ \_\|____|  |____|/_/ \_\|___/|___/

Overview

Appeal Labs is the consumer-facing arm of JPDEV, operating as a trading name. Most of what the lab builds runs headless: engines and pipelines with no interface at all. Anything built for people rather than for machines ships under this name.

What We Do

• Own Products: Building and launching our own B2B and B2C applications
• Operated, Not Delivered: We run what we launch, so hosting, maintenance and iteration all stay in-house
• Creator Partnerships: Collaborating with influencers and content creators on shared ventures

Partnership Model

We work with creators, influencers, and entrepreneurs who have audiences but need technical expertise. The structure is simple:

• Appeal Labs handles: All development, maintenance, hosting, and technical infrastructure
• Shared work: Design direction, theming, and content strategy are collaborative decisions
• Partner handles: Marketing, promotion, and audience engagement
• Revenue: Profit-sharing arrangement tailored to each project

This model lets creators launch apps without upfront development costs or technical overhead, while we gain access to established audiences and distribution channels.

Status

Active and open to new projects. If you have an app idea or an audience looking for a product, get in touch.

 __  __    _    ___  _  __ ___  _____    ___  ___ __   __  ___  _  _   ___   _      ___    ___ __   __   ___  _  _   ___  ___  _  _  ___
|  \/  |  /_\  | _ \| |/ /| __||_   _|  | _ \/ __|\ \ / / / __|| || | / _ \ | |    / _ \  / __|\ \ / /  | __|| \| | / __||_ _|| \| || __|
| |\/| | / _ \ |   /| ' < | _|   | |    |  _/\__ \ \ V / | (__ | __ || (_) || |__ | (_) || (_ | \ V /   | _| | .` || (_ | | | | .` || _|
|_|  |_|/_/ \_\|_|_\|_|\_\|___|  |_|    |_|  |___/  |_|   \___||_||_| \___/ |____| \___/  \___|  |_|    |___||_|\_| \___||___||_|\_||___|

Overview

Markets are moved by human psychology at scale. Fear, greed, euphoria, and panic drive price far more than fundamentals do in the short term. The Market Psychology Engine uses AI models to quantify that widespread sentiment and act on it before the crowd's mood is fully priced in.

How It Works

• Aggregates sentiment from news, social media, and market positioning data
• AI models score the crowd's mood, conviction, and direction
• Signals are expressed through leveraged positions or options structures to capture the move
• Position sizing and risk management are built for the volatility that comes with leverage

Why Sentiment

Fundamentals set the destination, but sentiment drives the path. When crowd psychology reaches extremes, prices detach from fair value, and those mispricings are exactly what this system is built to find and trade.

Status

Active and trading live. The engine is producing good returns so far, though with relatively high short-term volatility, which comes with the territory of leveraged and options-based positions. Ongoing work is focused on smoothing the equity curve without giving up the edge.

 ___  ___   ___  _  _    _    _       ___  _  _   ___  ___  ___  _____  ___   ___   _  _    ___  ___  ___  ___  _     ___  _  _  ___
/ __||_ _| / __|| \| |  /_\  | |     |_ _|| \| | / __|| __|/ __||_   _||_ _| / _ \ | \| |  | _ \|_ _|| _ \| __|| |   |_ _|| \| || __|
\__ \ | | | (_ || .` | / _ \ | |__    | | | .` || (_ || _| \__ \  | |   | | | (_) || .` |  |  _/ | | |  _/| _| | |__  | | | .` || _|
|___/|___| \___||_|\_|/_/ \_\|____|  |___||_|\_| \___||___||___/  |_|  |___| \___/ |_|\_|  |_|  |___||_|  |___||____||___||_|\_||___|

Overview

Every engine in the lab is only as good as what it gets fed. The ingestion pipeline is the layer underneath all of them: a set of scrapers and API collectors that run continuously, normalise what they find into a single schema, and score it for relevance before anything downstream touches it.

What It Collects

• Price and volume data across the instruments we trade
• Company filings and scheduled financial disclosures
• News wires and article text for NLP scoring
• Social and forum sentiment, weighted by account reach and history
• Macro and economic indicator releases

Design Notes

Sources break. Layouts change, endpoints get rate limited, and fields quietly change type. The pipeline is built to fail loudly rather than silently pass through bad data, because a corrupted feed that looks fine is far more expensive than a collector that stops and tells us.

Everything is timestamped at collection rather than at processing, so a strategy can be backtested against what was actually knowable at the time instead of what we know now.

Status

Running continuously and feeding the live engines. Current work is on widening coverage and improving how aggressively low-quality sources get down-weighted.

 ___ __  __ ___   ___  _   _  _____  ___   ___   _  _    ___  ___  ___  _  __   _       _   __   __ ___  ___
| __|\ \/ /| __| / __|| | | ||_   _||_ _| / _ \ | \| |  | _ \|_ _|/ __|| |/ /  | |     /_\  \ \ / /| __|| _ \
| _|  >  < | _| | (__ | |_| |  | |   | | | (_) || .` |  |   / | | \__ \| ' <   | |__  / _ \  \ V / | _| |   /
|___|/_/\_\|___| \___| \___/   |_|  |___| \___/ |_|\_|  |_|_\|___||___/|_|\_\  |____|/_/ \_\  |_|  |___||_|_\

Overview

A model deciding what to trade and a system actually getting that trade filled are two very different problems. This layer handles the second one, so the engines can stay focused on the first.

Responsibilities

• Position sizing: Translating a signal's conviction into an actual order size
• Exposure limits: Hard caps per position, per instrument and per strategy
• Order routing: Submitting to broker APIs and handling partial fills, rejections and retries
• Reconciliation: Continuously checking that what we think we hold matches what the broker says we hold
• Kill switches: Automatic shutdown on drawdown thresholds or unexpected position states

Why It Is Separate

Keeping execution independent of the engines means risk rules are enforced in one place rather than reimplemented inside every strategy. It also means a badly behaved model cannot talk its way past a limit, because the limit does not live inside the model.

Status

In development and being wired against live broker APIs. Running in a paper-trading configuration while reconciliation and the kill-switch logic are hardened.

 ___    _     ___  _  __ _____  ___  ___  _____    _  _    _    ___  _  _  ___  ___  ___
| _ )  /_\   / __|| |/ /|_   _|| __|/ __||_   _|  | || |  /_\  | _ \| \| || __|/ __|/ __|
| _ \ / _ \ | (__ | ' <   | |  | _| \__ \  | |    | __ | / _ \ |   /| .` || _| \__ \\__ \
|___//_/ \_\ \___||_|\_\  |_|  |___||___/  |_|    |_||_|/_/ \_\|_|_\|_|\_||___||___/|___/

Overview

Internal tooling, not a product. The harness is the gate every strategy has to clear before it is allowed anywhere near real capital, and it is deliberately built to be hostile to the strategies it tests.

What It Does

• Replays a model against historical data using only information available at each point in time
• Runs Monte Carlo simulations to check results against randomised outcomes
• Flags overfitting by comparing in-sample and out-of-sample behaviour
• Reports drawdown depth and duration, not just headline returns

On Trusting Backtests

A backtest is an argument, not a proof. It is easy to produce a beautiful equity curve by accident, which is why the harness reports the ways a strategy could be fooling us alongside the ways it looks good. A result that only survives on one time period, one instrument or one set of parameters gets treated as noise.

Status

In use across the trading engines. Ongoing work is on modelling transaction costs, slippage and liquidity constraints more honestly, since those are the assumptions most likely to flatter a result.

 _____  ___    _     ___  _  __   ___  ___   ___   ___   ___  ___
|_   _|| _ \  /_\   / __|| |/ /  | _ \| __| / __| / _ \ | _ \|   \
  | |  |   / / _ \ | (__ | ' <   |   /| _| | (__ | (_) ||   /| |) |
  |_|  |_|_\/_/ \_\ \___||_|\_\  |_|_\|___| \___| \___/ |_|_\|___/

Overview

The high-risk portfolio's year to date: the aggregate return, the options positions that drove it, and the individual trades that contributed most. Figures come from the engines' own execution records, include transaction costs, and are not audited.

Headline Numbers

• YTD return, high-risk portfolio, 4 September 2026: +379.01%
• ARK Innovation ETF (ARKK), YTD: +6.6%
• S&P 500 buy and hold, YTD: +13.2%
• Berkshire Hathaway (BRK.B), YTD: +5.1%

Versus The Field

• S&P 500 buy and hold: +13.2%
• Nasdaq Composite: +13.7%
• ARK Innovation ETF: +6.6%
• Berkshire Hathaway: +5.1%

Index and fund figures as of early August 2026. Different vehicles with different risk profiles, shown for scale rather than as peers: none of them run leveraged options books, and none of them drew down 34% in July either.

Monthly Detail

Six of the eight completed months were positive. Interactive Brokers YTD marks were +253.70% on 31 July, +370.42% on 31 August, and +379.01% on 4 September 2026. These imply an August return of +33.00% and a September-to-date return of +1.83%, calculated as (1 + ending YTD return) / (1 + starting YTD return) − 1. The best was January at +115.1%, the worst July at -34.3%. The two down months were not the same event: March was a broad selloff, with the S&P 500 down 4.3% and ARK Innovation down 4.7% alongside us, while July was specific to high-beta and leveraged positions, with ARK Innovation down 11.4% in a month the S&P finished flat.

Methodology

Only trades taken live count. Nothing is back-filled from backtests, transaction costs and slippage are included as executed, and losing periods stay in the record. The figures cover the high-risk portfolio only and are labelled as such rather than presented as the whole lab's return. This is also the strategy's first year live, and it trades infrequently, so the sample is small: a year of few large wins is not yet proof of edge, and we treat it accordingly. Past performance is not indicative of future results.

Status

Portfolio YTD figures as at 4 September 2026, from supplied Interactive Brokers screenshots. Benchmark figures remain as previously reported for early August 2026. The ledger updates when it is reconciled against broker records, not on a schedule.