Our production stack — the tools we train, deploy & serve with
We build production AI across speech, large language models, computer vision and GPU inference — for enterprise.
Our flagship product, QuantAI, applies that same institutional rigour to generate verified, out-of-sample options-trading signals.
Our production stack — the tools we train, deploy & serve with
Peer-reviewed academic research, Nvidia-certified GPU expertise, and a track record of shipping AI that hundreds of people use every day.
Trained like a research paper. Validated like a hedge fund. Every figure below is strictly out-of-sample, against real Alpaca OPRA market data.
Wyckoff market-structure regime detection · XGBoost long & short-horizon pair · DQN + PPO reinforcement-learning agents · Claude LLM real-time news sentiment · IV-aware options analytics (POP, Monte-Carlo CVaR, prob-touch-strike) — blended into a single confidence-weighted signal.
The DynamicStop engine adapts the trailing stop to volatility, sentiment, regime and time-decay — winners run, losers are cut early. A regime overlay scales sizing down up to 50% in adverse conditions, and a hard $1,000 capital cap protects every account. Leakage guards and cutoff validation keep the backtest honest.
Methodology: models are trained once with a hard cutoff date; all 52 test picks come from 8 rolling windows the model never saw — genuinely out-of-sample. The stock-only signal (no leverage, no options) carries a Sharpe of 3.44 — the edge exists before leverage even enters the picture. Past performance is not indicative of future results.
Research credentials and production delivery, in the same team. Every project below runs in a live environment.
A multi-model transcription service (Nvidia NeMo SALM, Nemotron-3.5 ASR, Meta Omnilingual ASR) served via Triton Inference Server on a GPU cluster. Dynamic model dispatch selects the best architecture per language and domain at inference time — delivered as a secure, multi-tenant service for large user bases.
LLM · RAG · AGENTIC
Custom LLM pipelines on open-weight models (GPT-OSS, Gemma, Llama, Qwen) and frontier cloud LLMs, with RAG architectures and 128K+ token context windows, deployed fully on-premise for sensitive data. The agentic layer chains ingestion, retrieval, reasoning and synthesis into a single secure query interface — no data leaves the perimeter.
VISION TRANSFORMERS · CLINICAL
A Vision Transformer (ViT) pipeline predicting outcomes from medical imaging. Baseline expert-level accuracy was ~40%; our model reached ~95%, and correctly recovered 3 of 4 mislabelled cases that human experts could not classify. Multi-GPU HPC training with full experiment tracking.
An Nvidia KAI scheduler with the Grove operator for efficient GPU-cluster deployment: gang-scheduling, dynamic resource allocation and priority management across training, inference and batch. Singularity/Apptainer for reproducible HPC alongside Docker for cloud-portable microservices, with AWS S3 and GCP for elastic burst.
An entropy-based optimisation method: +25% generalisation, −45% training time, +8% accuracy. Peer-reviewed in Pattern Recognition (IF 7.74).
Read the paper ↗Real-time hazard and object detection for ADAS from 3D LiDAR point clouds — pedestrians, vehicles and road obstacles segmented and tracked frame-to-frame for collision-avoidance pipelines.
Surface-anomaly detection at +23.1% accuracy via a custom CNN pipeline, deployed in a manufacturing QC line.
ML-driven liquid cooling: up to 62% power reduction, modelling millions in OPEX savings for US datacentres. IEEE-published and delivered as an EECA keynote (NZ Energy Efficiency & Conservation Authority).
Turn volatile data into actionable strategy. Our risk modeling suite delivers institutional-grade credit and mortgage default predictors alongside tailored insurance premium calculators. By blending robust risk assessments with advanced Transformer/RNN time-series forecasting, we’ve laid the groundwork for the QuantAI signal pipeline—giving your business a definitive predictive edge.
XGBoost + DQN/PPO + LLM sentiment + regime overlay. 63.5% OOS hit rate, Sharpe 4.65, +266% in 8 weeks. Real Alpaca data. Beta — pre-order open.
From multi-GPU training to real-time signal inference — one cohesive pipeline.
| Nvidia Stack | NeMo · Nemotron-3.5 ASR · Triton Inference Server · Dynamo · KAI scheduler + Grove · CUDA multi-GPU · cuDNN |
| Deep Learning | PyTorch · TensorFlow / Keras · Vision Transformers (ViT) · EfficientNet-V2 · Transformers · DQN / PPO RL |
| LLM / GenAI | GPT-OSS · Gemma · Llama · Qwen · frontier cloud LLMs · RAG · Agentic pipelines |
| MLOps / Infra | Kubernetes · Docker · Singularity/Apptainer · AWS S3 · GCP · Neptune.ai · Optuna · Labelbox |
| QuantAI-Specific | XGBoost · DuckDB · Polars · Alpaca OPRA · Tiger Broker · Streamlit · DynamicStop engine |
The technology is built. The track record is auditable. Consulting revenue funds operations while the SaaS scales.
| The Problem | Retail options traders lack the systematic edge of institutional desks — expensive data, no quantitative framework, emotional execution, zero risk management. The tools exist; they just cost $500K/year. |
| The Solution | QuantAI delivers a full institutional-grade autonomous trading system at SaaS pricing — the same ML stack as quant desks, the same rigorous OOS validation, accessible from $1,000 of capital. |
| Why This Team | Nvidia-certified DL experts (1 of 3 in ANZ), PhD-qualified, with 7+ years shipping production ML across clinical AI, enterprise LLMs, GPU inference at scale, and quantitative finance. |
| Market Size | US listed options set a record 15.2 billion contracts in 2025 (~61M/day, +26% YoY)1, with retail now ~50% of that flow.2 The algorithmic-trading market is forecast to reach $43B by 2030 (12.9% CAGR)3; AI consulting — our second revenue stream — $72.8B by 2030 (31.6% CAGR).4 |
| Revenue Model | Stream 1 — QuantAI SaaS: $99–$499/mo + fund licensing. Target 1,000 subscribers Year 1 = $1.2M ARR. Stream 2 — DL Consulting: existing enterprise AI revenue provides profitability while QuantAI scales. |
| Competitive Moat | (1) Verifiable walk-forward OOS track record · (2) Real options-bar data (Alpaca OPRA) · (3) Nvidia-stack inference · (4) PhD-level research discipline applied to risk. |
| Use of Funds | Productise the dashboard into a hosted multi-tenant web app · scale the data warehouse · go-to-market to quant/options communities · one full-stack hire. |
| Stage | MVP complete · paper-trading live · OOS results verified · real Alpaca data integrated — contact us for the full technical deck & P&L audit trail. |
Sources: 1 Cboe — “The State of the Options Industry: 2025” (15.2B contracts, ~61M/day, +26% YoY). 2 Fintech Global / Cboe — retail ≈50% of options volume (2025). 3 Grand View Research — Algorithmic Trading Market ($21.1B in 2024 → $43.0B by 2030, 12.9% CAGR). 4 Tech Market Experts — AI Consulting & Support Services Market ($14B in 2024 → $72.8B by 2030, 31.6% CAGR).
Whether you're an individual trader seeking an institutional edge, an enterprise needing bespoke DL infrastructure, or a VC evaluating a seed opportunity — we have the credentials, the code, and the data to back every claim.