Proof Of Work

Sanitized commercial analytics case studies first, backed by production build proof that shows the technical depth behind the work. Forecasting, pricing, margin, CRM, governance, and decision systems come before tool choice.

Use the filters to move between system builds and commercial analytics proof. The intent is to make the business relevance easy to scan, not to hide it behind categories.

Machine Learning 2024–Present Active

Algorithmic Trading AI System

A full-stack algorithmic trading ecosystem spanning three interconnected repositories (tf_1, Trading-Win, Trading-WSL) with strict ML/execution separation. The ML engine trains four model categories: hourly macro-regime classifiers (5-class directional labels STRONG_SELL→STRONG_BUY), high-frequency seconds-bar models with 30-minute lookback windows, multi-task L2 order book models predicting direction + bid-ask spread transitions + market impact regression, and a pre-market gap-opener ranker scoring probability of intraday moves from float, short ratio, relative volume, and news sentiment. Models are gated through a JSON-Logic promotion pipeline enforced via a shared ml-contracts schema repository (git submodule). The execution layer manages headless IBKR Gateway sessions (paper + live), appends market data to Parquet manifests, and consumes only models tagged production.alias. A shared FastAPI notification microservice (TelegramNotifications) delivers real-time alerts across the estate.

Key Highlights

  • Models: 4 model categories: hourly, seconds, L2 order book (multi-task), gap opener
  • Experiments: 19 W&B training campaigns (Nov 2025 – Feb 2026)
  • Infrastructure: 3-repo estate with shared contract governance via ml-contracts submodule
  • Data Ingestion: Databento L2 order book + IBKR historical bars in append-only Parquet manifests
TensorFlow / Keras scikit-learn W&B (Weights & Biases) Interactive Brokers API (ibapi) Databento (L2 order book) FastAPI Pandas / PyArrow / Parquet JSON Schema / JSON-Logic Python CUDA (GPU training) Docker pytest / CI

Build Proof

Illustration representing the AI Memoir & Narrative Pipeline project

Generative AI

2025–Present

AI Memoir & Narrative Pipeline

Deterministic GPT-4 narrative pipeline demonstrating retrieval-grounded generation, atomic patching, narrative safeguards, and auditable LLM workflow design.

Business relevance: Document intelligence, retrieval quality, narrative consistency, controlled LLM outputs, and audit-friendly workflow design.

GPT-4 (OpenAI API) FAISS (semantic search) sentence-transformers (all-MiniLM-L6-v2) spaCy 3.7 python-docx / pypandoc
  • FAISS-powered semantic retrieval over fragmented long-form source material
  • GPT-4 proposal generation with atomic chapter patch application
  • Narrative safeguards enforced on every pipeline run — not just at export
Illustration representing the AI Desktop Agent Orchestrator project

Automation

2025–Present

AI Desktop Agent Orchestrator

Python orchestration framework for multiple AI coding agents, demonstrating autonomous approval handling, prompt routing, transcript quality control, and operational cost monitoring.

Business relevance: Agent orchestration, process control, QA scoring, productivity measurement, and cost observability.

Python OpenAI Computer Use API pyautogui pyvda (virtual desktop management) uiautomation
  • OpenAI Computer Use API for visual button detection — no brittle pixel coordinates
  • Round-robin prompt distribution across simultaneous Copilot agent panels
  • Auto-generates fresh prompts from live project docs when agents go idle for 1 hour
Illustration representing the Multilingual Travel Authorization SaaS project

Full Stack

2025–Present

Multilingual Travel Authorization SaaS

Production-ready multilingual travel authorization SaaS proving regulated workflow design, LLM-assisted policy monitoring, and full-stack product delivery across payments, auth, and compliance.

Business relevance: Compliance workflows, policy monitoring, full-stack delivery, payments, authentication, accessibility, and production testing discipline.

Next.js 14 (App Router) TypeScript TailwindCSS PostgreSQL / Prisma Clerk (authentication)
  • LLM-powered automated scraping + analysis of government sites for policy change detection
  • Wizard-based multi-step application flows with family and group support
  • Full fee transparency compliance framework (government vs. service fee breakdown)

Sanitized Commercial Analytics Summary Cards

These are public-safe engagement summaries rather than standalone published case-study pages.

Summary Card • Forecasting & CRM Governance

Forecast Governance Reset

Public-safe engagement summary: forecasts were overstated or inconsistent because CRM behavior, stage discipline, and accountability varied across markets.

  • Diagnosed governance and behavior root causes before modeling
  • Improved review cadence and sales-cycle assumptions
  • Helped leadership move toward stronger forecast confidence

Summary Card • Power BI & KPI Governance

Commercial Analytics Foundation

Public-safe engagement summary from a broader foundation-fix brief: Sales, Finance, and leadership lacked one trusted view across revenue, margin, pricing, and opportunity data.

  • Standardized KPI definitions and improved reporting ownership
  • Supported centralized Power BI model and governance direction
  • Reduced conflicting narratives and improved decision speed

Commercial Leadership

Sales Engine From Zero

A new market required a functioning commercial engine under severe operating constraints.

  • Built targeting, distributor identification, and operating rhythm from scratch
  • Connected process, prioritization, and analytics to execution discipline
  • Delivered major growth improvement against the original target

Pricing & Margin Logic

Margin And Pricing Logic Repair

Pricing, FX, cost, rebate, and incentive logic distorted how leadership interpreted commercial performance.

  • Rebuilt decomposition logic and aligned methodology
  • Identified defects creating margin distortion across markets
  • Improved pricing and margin interpretation for leadership

Applied AI Enablement

Applied AI Enablement

Leaders wanted AI, but teams lacked shared language, governance, and practical use cases tied to real workflows.

  • Built training and enablement around practical business use cases
  • Connected governance, operating model, and adoption rather than hype
  • Positioned AI as a decision-system capability, not a side experiment

Start With The Health Check

If forecasting, pricing, margin, CRM, dashboard sprawl, or workflow prioritization are slowing the business down, the first step is to diagnose the decision system properly.