Securing corporate capital requires trust, not decorative design loops. This is a clean, hyper-focused portal for high-stakes capital advisory, paired with a specialized financial assistant trained to answer firm-specific questions without hallucinating data.
This one is in production for a working advisory practice. People arrive on it looking for capital, and the services it lists are the ones the firm actually sells: debt syndication, debt resolution, credit rating, virtual CFO and investment banking.
That changes what the work is. A personal project is finished when it runs. A site a business depends on is finished when someone who has never heard of the firm reads it, believes it, and knows who to contact. Everything below is in service of that.
Finance is the category where a visitor's first judgement is almost entirely visual. Someone weighing whether to hand their debt position to an advisory firm makes a decision about credibility in the first seconds, well before they have read a single service description.
So the design brief was narrower than it looks: not "make it modern", but "remove every signal that this is a small operation". Restraint over ornament, plain language over jargon, and one clear path from a service to a conversation.
I rebuilt it completely more than once to get there. Not iterated, rebuilt: each version taught me what was actually load bearing and what was decoration I had grown attached to, and the only way to find out was to throw the layout away and lay it out again.
The site carries an assistant called Finora. It opens by asking which area of the firm's expertise the visitor needs and offers the real ones: about the firm, the promoters, home and personal loans, business loans and syndication, credit rating improvement, insurance, startups.
The important decision there was the boundary. It answers for this firm only. A general purpose chat window on a financial site is a liability, because a plausible sentence about someone's loan eligibility is worse than no sentence, and a visitor cannot tell the difference between the two. Keeping it inside the firm's own material turns it from a novelty into what a receptionist does: work out what you came for and point you at it.
It also solves the real problem with an advisory site, which is that five services all sound adjacent to a visitor who does not work in finance. Asking one question routes them faster than a menu does.
A site nobody reaches is a design exercise. This one is indexed on purpose: an XML sitemap and robots file so the pages are crawlable, and page titles written for the services people search for rather than for the brand. The homepage title names the five services in order, because "Fenero Capital" is not what anyone types when they need debt syndication.
The logo is not mine. A friend designed it, and the palette and type on the site were built to sit with it rather than the other way round.
The site does its job and the firm is trading on it. The parts I would still argue with are the ones no visitor sees: it is a single page application where much of the content is static and would be better served rendered ahead of time, and the assistant's coverage is only as good as the material behind it, which means it needs revisiting whenever the services change. Neither is visible from the front. Both are the kind of thing that decides whether it is still good in a year.
When ChatGPT confidently lies to you. In casual chat, a made-up fact is funny. In corporate finance, a made-up loan interest rate is a lawsuit. We had to stop the AI from guessing.
Giving the AI an open-book test. Instead of letting the AI rely on its own memory (where it might hallucinate), we force it to read the firm's official documents and only answer questions based strictly on what's written on the page.