A guided investment workflow for MAG Capital, a lending application for Capital Access, hands-on AI enablement at Ascenda Talent, and Vagabod — a live product of my own. Each one says plainly what I built and the problem it was built to solve.
Guided investment workflow prototype
MAG Capital
Turning an investment framework into guided software
The situation. MAG Capital's Practical Portfolio Optimisation framework is its own — a considered method for how client goals, strategic and tactical allocation, risk and governance fit together. The framework didn't need rewriting. What it needed was software that would walk an adviser through it from the first conversation to the final report, so prospective institutional partners could watch it work rather than read about it in a deck.
What I built
A five-stage, fifteen-screen guided workflow — client and portfolio, strategy, risk and decisions, transition, presentation and review — so the method runs in order with nothing skipped
The calculation engine lifted straight from MAG's own workbook, so every figure is deterministic and traceable back to the framework — no number is ever generated by AI
AI where it earns its place: plain-language monitoring flags drawn from the same compliance checks the engine runs, draft adviser commentary and discussion reports, and presentation views picked from an approved chart library on a plain-English request
CIO model portfolios, an Approved Product List with monitoring and exception review, policy checks and MAG-branded reports
Outcome
A working prototype MAG can demonstrate end to end on synthetic client scenarios
Prospective partners can follow the reasoning behind every portfolio decision and see how adviser judgement and central oversight fit together
Why it matters
This is what AI in a governed investment workflow should look like — the AI drafts, flags and explains, the model does the maths, and the adviser keeps the final call
The same pattern — a specialist's framework in, guided software out — applies to any firm with a method worth productising
A demonstration prototype built for MAG Capital on synthetic client data. It isn't a bank deployment, and no endorsement by any institution is implied.
Turning lending expertise into a practical workspace
The situation. Preparing a private lending deal means keeping a lot of moving parts in agreement — security details, facility sizing, fees, each funder's contribution and a succession of documents — as it moves from indicative terms to confirmed funding. Capital Access already had the expertise and the calculations. What it needed was one place where a whole deal could be prepared, checked and progressed without anything drifting between the spreadsheet and the paperwork.
What I built
Mapped the lending process with the team and carried the source calculation logic into the application, checked against their own workbook
Debt sizing, loan-to-value and net-proceeds calculations, with individual funding contributions, rates and a funding reconciliation
A guided deal workflow that separates early assumptions from confirmed funding and keeps supporting evidence beside the step it belongs to
Document generation with previews and version history — indicative terms, offer material, instruction packs and lender notices — so what was prepared at each point is never silently overwritten
Outcome
A working application that brings the core components of deal preparation together, tested from intake to completion on synthetic lending scenarios
An active build — refinement continues as it beds into daily use, with the focus on simpler operation and practical document handling
Why it matters
Calculation logic that lived in a spreadsheet now runs inside a workflow, with the evidence and the documents attached to it
The same approach suits any firm whose deals move through stages and whose numbers have to stay consistent across every document that goes out the door
An active client build. Final documents and transaction decisions stay with Capital Access.
The situation. Executive search is a relationship business — the value is in the conversations, the judgement and the trust, and no client should ever feel there's an AI behind the interaction. Christian Mengler at Ascenda Talent wanted the preparation and the admin around those conversations taken off the plate, without handing over the decisions that make a search worth paying for.
What I did
Set up and configured the AI and meeting tools on Christian's own machine, connected the calendar and reviewed the privacy settings so candidate information isn't used to train anyone's models
Tested voice input so dictation replaces typing, and showed how a recorded meeting becomes a structured recap and a drafted follow-up
Encoded that post-meeting workflow as a reusable skill, so it runs the same way every time without being rebuilt from scratch
Demonstrated a candidate report that brings a CV, interview notes and the brief criteria together for review — surfacing the facts and the gaps so the consultant makes the call
Outcome
A configured starting point and concrete, repeatable examples Christian can run without me in the room
A clear line between what to automate — brief synthesis, notes, report assembly, follow-ups — and what stays human: the calls, the cultural fit and the candidate decisions
Why it matters
Most firms don't need a platform first. They need the tools set up properly on their own machines, with someone showing them what good looks like
Privacy and human review were built in from day one, which is the only way AI belongs in a relationship business
AI advisory, tool configuration and workflow demonstrations — not a delivered recruitment platform.
The situation. Vagabod is a real live site that helps travellers find casual gym access anywhere in the world — it solves the problem of gyms not advertising casual membership without a traveller enquiring first. My team of agents reaches out to gyms around the world in multiple languages and enriches a database with information not available from a simple google search or a gyms website.
The live Vagabod landing page — an enriched directory of gyms with casual access across dozens of countries.
What I built
Automated scraping + enrichment pipeline
Casual-membership pricing extraction per gym
Outreach pipeline to inquire about unadvertised casual offerings
Continuous re-enrichment as data drifts
Outcome
7,986 gyms across 59 countries enriched
Directory kept fresh without manual entry
Self-service claim work flow for gym owners to keep their pages current
Why it matters
The same agentic enrichment applies to any large, complex dataset
over 10 Automated workflows & always on, running 24/7 on local LLMs hosted on private infrastructure, for free
