@Jeuri

AI/ML engineer & BBIT grad 🇰🇪 · AI tools that save time & money · Scam safety · AI videos · Arsenal ⚽ / 📩 DMs open for AI, ML, data & video work
🤦

[@Jeuri @Jeuri2406]

Approval first. Nothing important runs until you say yes.

I'm building Nexuss: an AI assistant that shows you its plan, waits for your tap, then gives you a receipt for every step it took.

Building in public. What would you trust an AI to do for you?

[@Jeuri @Jeuri2406]

@TheAbojani Before any IPO masterclass Q&A, write down one use-of-proceeds question and one valuation question. If less than half the raise funds growth (stores, inventory, tech) vs early exits, size the buy smaller than your gut wants.

[@Jeuri @Jeuri2406]

@CNBC Bond yields near 15% on AI borrowing is a cash-flow warning, not just a stock move. Quick check: if your own AI tools bill would still be worth it at 2x the subscription, keep them; if not, cut the nice-to-haves first.

[@Jeuri @Jeuri2406]

@BrianFeroldi The "appropriate time period" is the whole point. Rule I use: ignore any return chart shorter than 10 years when someone is selling a wealth story. One-year and even 3-year windows flip too often to plan around.

[@Jeuri @Jeuri2406]

@aakashgupta The checkmark-with-no-words trap is real. Fix that works: paste the pricing table as text, then ask the model to answer only by quoting the exact plan name and the cell under that feature. No paraphrase allowed.

[@Jeuri @Jeuri2406]

@CNBCMakeIt If they ask you to draft your own recommendation letter, lead with 3 numbers they can stand behind: a metric you moved, a project size, and a timeframe. Managers sign faster when the proof is already written for them.

[@Jeuri @Jeuri2406]

@kenyanwalstreet Taxes at 33.2% of that KSh 226.96 petrol price is the part that doesn't move when landing costs dip. Habit that helps: when EPRA posts, jot the tax line separately from the pump price so you don't blame the whole jump on crude.

[@Jeuri @Jeuri2406]

@TheAthleticFC @tombogert For U22 rankings I'd weight league minutes over highlight moments. A 19-year-old with 2,000+ MLS minutes and steady output is a far safer bet to make the jump than one who looks great in 20-minute cameos off the bench.

[@Jeuri @Jeuri2406]

@Arsenal Proper squad selfie. Phone tip if you're snapping at the Emirates: tap your face on screen before you shoot so exposure locks on you, not the floodlights behind. Saves you from the classic dark-silhouette matchday pic.

[@Jeuri @Jeuri2406]

@svpino Reproducibility is the part most agent tools skip. Habit that saved me on my football models: log package versions and the random seed with every run. A pandas or sklearn bump alone can shift a result enough that you can't rerun your own old number.

[@Jeuri @Jeuri2406]

@everydayimruss Yeah — speak the ask, read the plan. Names and amounts stay on screen so you can spot-check without replaying the whole thing. That’s the approval step I’m trying to keep light.

[@Jeuri @Jeuri2406]

Type or speak. See the exact plan. Then approve.

I'm building Nexuss in public: an early prototype of an approval-first personal AI assistant. You ask (type or talk), it shows the exact plan, and nothing important runs until you tap yes.

Benefit I'm designing for: you see what would happen before it happens.

Would you rather type or talk to an assistant that waits for your yes?

[@Jeuri @Jeuri2406]

Extra check I use now: train only on matches before a cutoff date, score only on matches after it, then recalibrate (I used isotonic) on a middle window so confidence stays honest on new fixtures.

[@Jeuri @Jeuri2406]

Shuffling match data lied to me.

On 4,258 games my football models looked solid: 0.952 log loss with a random split. Same models, cut by date instead: 0.981.

Future form was leaking into the past. For any time-series sports model, split by calendar — never shuffle.

How do you stop leakage when you train on seasons?

[@Jeuri @Jeuri2406]

@CNBCMakeIt The part people skip with 3D-print hustles is unit cost. Before buying a second printer, log filament, print hours and failed prints per item for your first 20 orders. If an item doesn't clear about 3x its material cost, drop it or reprice it before scaling.

[@Jeuri @Jeuri2406]

Shuffling match data lied to me.

On 4,258 games my football models looked solid: 0.952 log loss with a random split. Same models, cut by date instead: 0.981.

Future form was leaking into the past. For any time-series sports model, split by calendar — never shuffle.

How do you stop leakage when you train on seasons?

[@Jeuri @Jeuri2406]

@YahooFinance Cuts right before earnings are usually about the margin story. If you're in tech or media, do a 30-minute check this week: save your best work samples outside company tools and count how many months of expenses your savings cover. Under 3 is the first thing to fix.

[@Jeuri @Jeuri2406]

@kenyanwalstreet That lag is worth planning around. EPRA's monthly review comes out on the 14th, so if you drive for work, fill up on the 13th or 14th when an increase looks likely. A full tank at the old price buys you a week or more before the new rate bites.

[@Jeuri @Jeuri2406]

@aakashgupta The CRM math is the scary part. If a rep makes ~200 touches a day and an agent makes 20x that, give each agent its own API key with a hard daily write cap and queue its updates in batches. Then one runaway agent can't take the system down for the humans.

[@Jeuri @Jeuri2406]

@mohammad_saaaad Motion design is my thing — promo animations, logo stings, clean transitions. Happy to help: what's the motion for, roughly how long, when do you need it, and do you have a budget in mind?