What AI Implementation
Actually Looks Like

Real projects. Real results. Not demos — deployed solutions with measured before-and-after outcomes.

Financial Intelligence

From Spreadsheet to Executive Dashboard — in One Week

A manufacturing company was managing $3M+ in annual revenue from a P&L spreadsheet with no way to see trends, spot problems, or brief the ownership team at a glance. We turned their raw Excel data into a live, interactive financial dashboard — deployed and password-gated in under a week.

✓ $718K profit swing visible at a glance — from −$403K to +$315K YoY

The Problem

The company had real financial momentum — $3.42M in YTD revenue, improving margins, and a 46% drop in overhead per cubic metre — but none of it was visible in a format leadership could act on. Every answer required digging through spreadsheet tabs. There was no single view of what was working, what had changed, or where costs were moving. The owner couldn't brief investors or staff without manually compiling numbers each time.

What We Built

  • Parsed and structured 11 months of P&L data from raw Excel into a clean JavaScript data layer
  • Built a 5-tab interactive HTML dashboard: P&L Overview, Overhead & Volume, Revenue Mix, Cost Breakdown, and Quarterly Summary
  • Surfaced 6 headline KPI cards including YTD revenue, gross margin, net profit swing, overhead per m³, and volume delivered
  • Added monthly trend charts, quarter-over-quarter comparisons, and a full line-item P&L table with prior-year benchmarks
  • Deployed to Netlify with a password gate — shareable with lenders, advisors, or ownership in seconds

Service Used

AI Opportunity Audit + Implementation Sprint

Infrastructure

AI Working Environment Build

From generic AI outputs to an environment that starts every conversation at 80%. Built a custom AI operating environment that eliminated cold-start prompting and produced consistent, high-quality outputs from day one.

✓ Consistent outputs from first prompt

The Problem

AI outputs were inconsistent and required heavy editing. Too much time was spent re-explaining context. Every session started from zero.

What We Built

  • Custom system prompts per work type and context
  • Persistent context files for AI to load at session start
  • Output templates for common deliverable types
  • Prompt library for the team's top 10 use cases
  • Training session on context architecture

Service Used

AI Opportunity Audit + Implementation Sprint

AI Automation

CommandOS — A Fully Capable AI Agent in Your Pocket

We built an always-on Telegram bot that gives our founder complete access to Claude AI from any phone, anywhere. Text it, voice-note it, send a photo — it thinks, searches, writes, generates PDFs, and delivers results back in seconds.

✓ Full AI agent access from any phone, 24/7

The Problem

A founder's best thinking doesn't happen at a desk. Ideas surface in the car, decisions get made on the go, and waiting until you're back at a laptop means losing the moment. Off-the-shelf tools couldn't read our files, run our workflows, or operate within our business context.

What We Built

  • Persistent Claude agent connected to a private Telegram group — always primed with full business context and workspace access
  • Voice note transcription via Whisper — speak your request, get a typed-and-actioned response
  • Photo and screenshot analysis — send an image, the agent reads and interprets it
  • Automatic file delivery — PDFs, charts, and research outputs pushed back to Telegram as attachments
  • 24/7 uptime via macOS launchd — auto-restarts on reboot, posts a boot confirmation to Telegram on startup

Built For

TFS Internal — we built this for ourselves. We can build it for you.

AI Systems

Multi-Agent CFO Q&A for a 6-Entity Holding Group

Built a fleet of specialized AI agents that answer financial questions across a complex corporate group — instantly, with every number traced to a source document. What used to take hours of manual document search now takes seconds.

✓ 21 source documents · every answer cited · 13-question validation suite

The Problem

The leadership team of a 6-entity holding group needed fast, reliable answers to financial questions spanning multiple subsidiaries. The answers lived across dozens of PDFs: financial statements, audits, loan documents. Finding them meant searching manually, cross-referencing across entities, and hoping nothing was missed.

What We Built

  • 6 specialized AI agents — one Group CFO orchestrator plus one agent per entity — each trained on its own financial knowledge base
  • Knowledge base built from 21 source PDFs with full citation tracing back to the original document
  • 13-question validation suite to verify every answer references a real source before deployment
  • Deployed as slash commands — usable directly from the team's existing workflow
  • Phase 2 roadmap: live accounting integration to connect AI answers directly to real-time bookkeeping data

Service Used

AI Implementation Sprint

Investor Relations

Investor Brief for a Multi-Million Dollar Energy Capital Raise

Built a production-ready investor brief for an energy company pursuing European capital — delivered in 2 weeks, active in investor conversations immediately after. Every number traces back to a single financial model so updates take minutes, not days.

✓ Active in investor conversations within 2 weeks of first brief

The Problem

The company had a compelling story — audited operating assets, a committed build pipeline, and 26 sites of greenfield opportunity — but no document that could hold up to institutional scrutiny. Financials lived in a spreadsheet. Narrative lived in the founder's head. There was no way to update numbers quickly across a multi-section brief without manual errors creeping in.

What We Built

  • 8-section executive investor brief in HTML and PDF — password-gated, with sticky navigation and a three-layer revenue model
  • Python regeneration pipeline that pulls every figure directly from the canonical financial model — change one cell, regenerate the brief in seconds
  • SVG asset register visualization and cumulative cashflow chart built to institutional presentation standards
  • Sensitivity heatmap showing DSCR and EBITDA across revenue and cost scenarios — gave investors the stress-test view without a separate model
  • 8 versions delivered across 2 weeks of live iteration, finishing with a 5-year ramp to ~$4.8M revenue and Y5 EBITDA of ~$2.2M at 2.98× DSCR

Service Used

AI Implementation Sprint

Team Enablement

AI-First Mindset Workshop

Shifted a consulting team from using AI as a search box to managing it like a capable team member. Not a tool demo — a structured half-day session that changed how the team thought about their own work.

✓ Team AI usage up within 2 weeks

The Problem

The team had access to AI tools but adoption was flat. People used ChatGPT occasionally for search-like queries and went back to their normal workflows within a week.

What We Built

  • Half-day workshop — no slides, all hands-on
  • Live workflow mapping of real team tasks
  • Each person identified their own top 3 use cases
  • 30-day implementation plan per person
  • Follow-up accountability check-in at week 2

Service Used

AI Transformation Coaching (workshop format)

Content Operations

YouTube Content Workflow Automation

From 12+ hours of manual work per video to a streamlined, AI-assisted production system. Every step — from ideation to publish — was mapped, automated where possible, and handed off with documentation.

✓ 12+ hours recovered per video

The Problem

A content creator was spending the majority of production time on low-value work: researching topics, writing scripts from scratch, creating thumbnails, and writing descriptions — all manually, every time.

What We Built

  • Automated ideation pipeline pulling from research sources
  • AI-assisted script generation from outline to draft
  • Thumbnail brief generation based on video topic
  • Auto-generated descriptions, tags, and chapter markers
  • End-to-end workflow with human review checkpoints

Service Used

AI Implementation Sprint

AI Automation

CommandOS — A Fully Capable AI Agent in Your Pocket

We built an always-on Telegram bot that gives our founder complete access to Claude AI from any phone, anywhere. Text it, voice-note it, send a photo — it thinks, searches, writes, generates PDFs, and delivers results back in seconds.

✓ Full AI agent access from any phone, 24/7

The Problem

A founder's best thinking doesn't happen at a desk. Ideas surface in the car, decisions get made on the go, and waiting until you're back at a laptop means losing the moment. Off-the-shelf tools couldn't read our files, run our workflows, or operate within our business context.

What We Built

  • Persistent Claude agent connected to a private Telegram group — always primed with full business context and workspace access
  • Voice note transcription via Whisper — speak your request, get a typed-and-actioned response
  • Photo and screenshot analysis — send an image, the agent reads and interprets it
  • Automatic file delivery — PDFs, charts, and research outputs pushed back to Telegram as attachments
  • 24/7 uptime via macOS launchd — auto-restarts on reboot, posts a boot confirmation to Telegram on startup

Built For

TFS Internal — we built this for ourselves. We can build it for you.

Business Intelligence

DataOS — Automated Business Intelligence Pipeline

We built our own daily intelligence layer so our AI assistant always has current business data — no dashboards to open, no reports to pull. Every morning at 6 AM, the pipeline runs itself.

✓ Daily metrics, zero manual effort

The Problem

Every AI-assisted session started from stale data. Business metrics lived in disconnected platforms — YouTube Analytics, Google Analytics, FX rates — with no automated way to surface them. Getting a current picture meant logging into three tools and doing the math manually, every time.

What We Built

  • Automated collection pipeline running daily at 6 AM via launchd — no cron, no cloud, no maintenance
  • Auto-discovery architecture — drop a new collect script and it runs automatically, no wiring required
  • SQLite data warehouse aggregating YouTube Analytics, Google Analytics 4, and live FX rates
  • Auto-generated metrics snapshot loaded into every AI session at startup
  • Daily Brief integration — fresh intelligence delivered to Telegram every morning before the workday starts

Built For

TFS Internal — we built this for ourselves. We can build it for you.

Operations Automation

Daily Brief — Automated Morning Intelligence

Every morning at 7 AM, an AI-synthesized briefing lands in Telegram before the workday starts. Business metrics, research highlights, and key priorities — compiled automatically, no manual effort required.

✓ Full business snapshot delivered every morning at 7 AM

The Problem

Staying on top of a business means checking multiple sources every morning — metrics, research feeds, priorities, meeting follow-ups. That review either doesn't happen consistently, or it burns 30–45 minutes before real work begins. Most founders don't have a chief of staff to do it for them.

What We Built

  • SQLite-backed pipeline that collects business metrics daily at 6 AM from connected sources (YouTube, Google Analytics, FX rates)
  • Gemini AI synthesis engine that pulls metrics, business context, meeting transcripts, and activity into a single structured briefing
  • Configurable presets so the briefing format adapts to the operator reading it
  • Dashboard image generator that visualizes the funnel snapshot and attaches it directly to the Telegram message
  • Fully automated schedule — runs at 7 AM, saves to file, delivers to Telegram. No trigger, no login, no manual step

Built For

TFS Internal — we built this for ourselves. We can build it for you.

Personal Productivity

Bespoke Meal Planner

A fully automated, personalized meal planning system — from macro targets to grocery list. Built to run weekly without manual input, adapting to preferences, what's in the fridge, and what's on sale.

✓ Zero manual weekly meal planning

The Problem

Weekly meal planning was consuming 45–60 minutes and still producing inconsistent results — mismatched macros, forgotten groceries, decision fatigue by Wednesday.

What We Built

  • Macro-aware weekly meal plan generator
  • Recipe database integrated with nutritional data
  • Automated grocery list sorted by store section
  • Preference engine that learns over time
  • One-click regeneration for any meal

Service Used

AI Implementation Sprint

Want results like these?

Every project starts with a conversation. Tell me what's slowing your team down and we'll figure out if there's a fit.

Book a Conversation