We design and build the products behind customer conversations, conversation intelligence, meeting capture and the operations software underneath them — from the model layer down to the queue that runs it at 3am.
An agent that answers. A person who can take over.
Connect WhatsApp, Messenger and Instagram to one inbox and put an AI agent behind it — trained on your own site, documents, Q&A and catalogue, not on a generic script. It replies, tags, books and takes orders on its own, and hands the conversation to a human the moment someone clicks Takeover.
Book a walkthroughbotista.io ↗
Every call your team takes, read and scored.
Recordings land, get transcribed and split by speaker, then scored for risk, sentiment and the coaching moments a manager would otherwise never hear. Each call answers the five questions a sales lead actually asks — what they wanted, who decides, the budget, the timeline, the outcome — with a timestamp on every claim.
Book a walkthroughcall-lens-analysis.vercel.app ↗
Speak your language. The meeting hears theirs.
A desktop app that translates your voice while you talk. It speaks the translation into a virtual microphone, so Zoom, Teams or Meet hear it as an ordinary mic — nothing to install inside the meeting app. It runs the other way too, and the keys stay encrypted on your own machine.
Book a walkthroughmeet.botista.io ↗
Run the venue, not the paperwork.
The all-in-one platform for wedding halls and event venues — bookings, contracts, payments and a customer microsite per venue, in one calm workspace. English and Arabic out of the box, because half the market works right to left.
Book a walkthroughplanniyo.com ↗
Every meeting, resolved into decisions and owners.
Capture on mobile or desktop. Notes transcribes, attributes each turn to a speaker, and returns the decisions, owners and open questions — with a timestamp on every claim so you can jump back to the moment it was said.
Book a walkthroughThe model is the easy part. We build the system around it that keeps working when the model doesn't.
Ingest, transcode, diarise, transcribe, align. Pipelines that don't drop a frame at 3am.
Tool-using systems with guardrails, human checkpoints, and an audit trail you can hand to a regulator.
Applications built to be handed over, not babysat. Your team should be able to read the code.
Queues, workers, storage, observability. The part that decides whether it survives Monday.
Team work, on other people’s products. We architected, led or built these — we don’t own them.
An enterprise sales platform built on a broker hierarchy: leads distributed by rotation, then contracts, commissions and renewals tracked through to payment, with activity history and multi-channel notification underneath.
The lead lifecycle end to end — calls, contacts, statuses and quotations compared across providers — with bulk sheet import and validation, a do-not-call list, and push, SMS and WhatsApp notification.
A two-sided app for staffing events. Clients brief an event down to dress code and languages and are matched with ushers; ushers confirm, check in and out on site by QR and location, and are paid out through an in-app wallet. An admin dashboard sits behind it for staff profiles, event dates and payouts.
A multi-tenant platform for universities: courses, semesters, assignments, quizzes and weighted grading, alongside student portfolios, discussion forums, announcements and attendance.
Short-stay rentals with the operational parts that decide whether they work: availability reconciled against external booking channels, multi-currency rate plans, promo codes, broker commissions and scheduled guest email.
Multi-tenant digital menus for restaurant groups — brands, branches, categories and items, each localised in English and Arabic at the table level rather than bolted on — with customer reviews and questionnaires on top.
Padel court booking with overlap-free slot validation against working hours and time-offs, an in-app wallet running alongside cash and card, and an admin side for courts, events and payments.
An agricultural advisory chatbot inside an e-commerce platform: crop and disease questions asked as text, image, audio or PDF, answered in Egyptian Arabic from a vector search over the company's own agronomy material, with the right product attached to the answer.
A personal assistant that answers the phone for you — speech streaming in and out, calls routed by intent, spam filtered, and what can be booked booked — over a telephony backend carrying push, analytics and in-app purchases.
An enterprise agent running inbound and outbound calls through a PBX, with an LLM, speech-to-text and text-to-speech behind a queue-driven multi-tenant backend.
A subscription service that reads WhatsApp in real time, classifies spam and archives it — several connected accounts per plan, with classification done asynchronously and results broadcast back live.
Buy-in-installments over a multi-tenant storefront: catalogue synced from Shopify, payment through Paymob, shipping through Bosta, and the whole order, delivery and refund lifecycle driven by webhooks.
Property listings with VIP placement and per-listing visitor analytics, multi-language from the schema up rather than through a translation layer, and sign-in by password or one-time code.
Car rental for the Saudi market, booked from the phone: pick a branch or have the car delivered, set the dates and times, then choose the car itself. Arabic and English are a first-run choice rather than a setting buried later.
The second rental app for the same operator, built around the whole journey rather than the booking alone — search and reserve, pick the car up against a QR code, then tracking, insurance cover and reporting once it is on the road.
Buying at wholesale from a phone: browse by category, add every variant of a line to one order, review the whole basket before committing, and see the running offers. Arabic-first throughout, because its buyers are.
Botista exists for the second half of that sentence. We are a small studio of engineers and designers who take AI systems the rest of the way — through the queue backlogs, the provider outages, and the four-hour recording that breaks the parser.
Anything looks good on the happy path. We design for the timeout, the malformed audio, the provider that starts returning nonsense at four times the usual latency.
If a person can't see what the system decided and why, the model's accuracy is beside the point. Every inference we ship is traceable back to the moment it came from.
We write the code your team will inherit. No private conventions, no undocumented magic, no permanent dependency on the studio that built it.
We find the constraint that actually decides the shape of the product. Usually it is not the one in the brief.
Interface and architecture designed together. A data model that can't represent an invalid state, and a screen that shows why.
Weekly builds you can use, not a demo at the end. Every release runs against production-shaped data.
Instrumented before it ships. We know what broke before your users tell you, and we know what to roll back to.
We stay on for the part most studios skip: the months where load, cost and edge cases show you what you actually built.
Tell us what you are building and where it is stuck. We reply to every message ourselves, usually within a day.