I looked inside the platform, set the direction, and drew the first screens. This is the proposal to review, approve, or mark up.
The trade model is right and the reference data is there. The limits were structural — and every one is fixable in the rebuild.
Offers, requests, varieties, payment terms, Incoterms, transport by sea / air / rail / road — the trader's language is captured. Kept and carried forward, not re-invented.
A match was calculated by a scheduled job and waited in a table until someone signed in. For perishable goods, the match must travel to the trader — in seconds.
Produce fields were hardwired as fixed columns. I fix this once, now, so flowers or a dry commodity drop in as configuration — never a rebuild.
Every inner-page heading sat on a banner hot-linked to a host that went dark — white on white. One bug, whole-site symptom — the kind of thing a clean rebuild simply doesn’t carry.
Privacy Policy and Terms were dead links; security headers absent — on a platform holding company names and numbers across borders. Built in properly from day one on the new platform.
No analytics of any kind. The rebuild instruments everything from day one, so decisions run on evidence, not guesswork.
The honest headline: the platform is early — reference lists loaded, real trading only just beginning. Which is the right moment to fix the foundations, because nothing has to be unwound. I’m building the new platform from scratch — a clean rebuild — while the current site keeps running until the new one replaces it.
I drew these on Hookipa's own warm identity — the seedling mark, the slab wordmark, leaf-green and cream. Real screens now, not sketches. Yours to approve or redirect.
| Product | Grade | Origin | Qty | Price ≈/kg | Match |
|---|---|---|---|---|---|
Navel oranges | Class 1 | W. Cape | 24 t | R 8.20 | 92% |
Valencia oranges | Class 1 | E. Cape | 40 t | R 7.10 | 88% |
Table grapes | Class 1 | Orange R. | 12 t | R 41.80 ⚠ high | 71% |
Hass avocados | Count 18 | Limpopo | 6 t | R 38.50 | 84% |
Still to come (next pass): the platform-admin taxonomy screen in full, the posting flow, every empty / loading / error state, French & Spanish layout checks, and the reusable component library. None changes the direction you're approving here.
The design settles how people are organised — individuals act for a company, and support can stand in any of their shoes to help.
Posts offers, works matches.
Posts requests, reveals sellers.
Team, verification, billing.
Multi-commodity taxonomy, oversight.
Views & acts as any role to help.
Offers belong to the company, not the person. A company can invite colleagues, and a support "view-as" lets me stand in a member's shoes to resolve a problem — pinned at the top of the workspace and clearly marked, never hidden.
AI works hardest when nobody notices it. It never leads the pitch — it removes friction, catches mistakes, and crosses the language barrier, while a person always stays in control.
Forward an availability email or drop a spreadsheet — it returns structured offers to confirm. A five-minute chore becomes one click, lifting supply.
An offer well off the 30-day average is flagged with the reason — protecting buyers, keeping the market honest.
A French offer reads in Spanish for the buyer, English for the next — original always one tap away.
Each match carries a score and a plain-language reason — so a trader trusts it enough to pay for the reveal.
The rule, always: nothing an AI produces is published, sent or shown until a person approves it. Every model call is logged with its cost, so spend is visible to the rand. AI is a feature of the plumbing — never the headline.
Defined items across 13 groups, each with a plain test that proves it works. Phased so value lands early and you decide, at each edge, whether to continue. The items marked new came out of the design's first pass.
Everything today's site does, rebuilt — plus the five roles, a search that scales, matching that runs on write, and a taxonomy ready for more than fruit.
60-second alerts by push, WhatsApp & email; company verification; pay-to-view reveals, bundles and billing.
Signed iOS & Android apps, iPad, Apple Watch & Wear OS — the match in the trader's pocket and on their wrist.
This is the shape. The full item-by-item scope — 116 checks, each with a plain test that proves it’s done — is kept for your team and opened via a login, not scattered through this page. A few items still need your decision (gathered in the sign-off below); the partnership terms are agreed together before I start.
Colour and type come from the seedling mark and slab wordmark — leaf-green, cream, and a warm serif that carries the headings and the prices. So the rebuild reads as the same brand, matured.
One rule I hold to, learned from the live bug: a heading always sits on solid colour, never a hot-linked image. If a picture fails tomorrow, the page still reads perfectly.
Same build, same quality. What changes is how you invest and how much you carry up front. I’m comfortable with any of the three.
I carry the build and back it with my own capital; you put in a modest, fair contribution rather than the full cost, and I share the upside as genuine partners. Lowest cash out for you — and I do well only when the platform does.
Prefer to own it outright with no shared upside? Commission the whole build in one go — simplest, and the best value per rand.
Or pay for one phase at a time and decide at each boundary whether to continue. Lowest risk — every step proves its value before the next.
What you get, whichever route: a platform built on a proven kit that turns matches into revenue and reaches traders on web, phone and watch — with the trade knowledge from the existing system carried across, not rebuilt. Under the partnership you own it too, with the terms set out in a short, fair agreement I sign before starting.
My recommendation: the fair partnership. It gets the platform built without a heavy cash outlay from you, keeps both of me pulling in the same direction, and I agree the split and a short term sheet are agreed together before a line of it is built.
Tick what you're happy with, mark up anything you're not. Nothing is locked until you say so — and the name stays a setting I can change late.
Mark up this page, reply on WhatsApp, or note it in my next call. Each change is costed transparently against the agreed scope, so you always see what it adds or saves before it's done. Once live, "Report a problem / request a change" sits inside the product.
How much a member sees before paying; how many free reveals a newcomer gets; whether an offer leaves the market once a reveal is bought, or stays open to several buyers; bundle sizes; card-only vs EFT for larger customers; which markets beyond fresh produce, and in what order; and the name itself — it must be findable, so I settle it before any artwork or store listing.
A short, plain term sheet — what each side puts in, how I share the upside, and the gates along the way. Signed before I start.
I agree the open decisions above and lock the build list, so both sides know exactly what’s being built.
The rebuild, then the service layer that earns revenue, then the mobile and wearable reach — each phase a clean milestone.