The Asher Suite
Asher is a family of production platforms spanning market data, charting, artificial intelligence, commerce, and speech. Kevin Hokoana is its architect — the data model, the engine, the interface, the agent, and the business model underneath it.
It exists on this site for one reason: it is the evidence. Every claim the consulting practice makes about AI systems, data infrastructure, and shipping software was learned here.
Status is stated plainly. Live means running in production today.
Trading platform
The flagship platform — terminal, strategy tester, market data, accounts, and billing under one application.
A purpose-built strategy language, a deterministic backtest engine that runs in the browser, twenty timeframes, and per-chart configuration. Every chart is drawn by a first-party engine written from scratch — nine chart types and roughly fifty drawing tools.
Consumer charting
A trader-first charting terminal for crypto, equities, and FX — the consumer face of the same engine.
Indicators, drawing tools, watchlists, alerts, saved layouts, and bar replay. A hundred and fifteen studies in a categorised picker. Built as a direct alternative to the incumbent platforms, with the analysis surface as the product rather than an embed widget bolted onto someone else's stack.
Market data API
The market-data layer, packaged and sold as its own product — data as a service.
Candles, live prices, funding, order books, and an instrument catalog across crypto, equities, and FX. Ingested from five venues, normalised into a single store, and served through a read-through cache with scheduled backfill.
AI assistant
A conversational agent wired to live platform data, with voice, real tool use, and a curated knowledge base.
A real tool-use loop across eight tools — not a scripted chatbot. Prompt caching and inference-free fast paths hold the per-conversation cost down; speech in, synthesised voice out, with barge-in handling for natural conversation.
Multi-tenant SaaS
A help desk with tickets, email, scheduling, and per-tenant isolation — sold to other companies as software.
Multi-tenant from the database up: row-level security across every domain table, proven by an automated isolation test. Inbound and outbound email per tenant, scheduling built to category parity, and a marketing surface shipped in seventeen locales.
Speech infrastructure
Self-hosted speech synthesis and transcription — the voice layer, built rather than rented.
A GPU inference stack that took synthesis latency from sixteen seconds to a warm median near one, with a self-trained wake-word model and an ordered provider fallback chain. Vendor choice treated as an engineering variable rather than a lock-in.
Six layers, all first-party. Each one was a decision to build rather than rent, and each decision is defensible on its own terms.
Candles ingested from multiple venues, normalised into a single store, served through a read-through cache with scheduled backfill and gap repair. Nothing in the stack depends on a third party's chart feed — which means no borrowed outages, no licence ceiling, and no pricing power held by someone else.
A purpose-built strategy language and a deterministic backtest engine. Identical results whether a strategy runs on the server or in a browser worker — because a backtest you cannot reproduce is a story, not a result.
The chart renderer is first-party code, not a licensed widget. That decision cost more up front and bought complete control over the surface — custom overlays, risk zones, live price and countdown, bar replay, and saved layouts that behave exactly as designed.
A conversational agent that calls tools against first-party data rather than improvising. Engineered for cost as much as capability: prompt caching, zero-inference fast paths for common questions, and a curated knowledge base so the model retrieves instead of guessing. Speech in, synthesised voice out, with barge-in.
Embedded checkout on the platform's own domain, fail-closed webhook handling, a usage meter, and an entitlement model that gates capability against what a customer has actually paid for. The unglamorous layer that decides whether software is a product or a demo.
The execution layer signs orders in the user's own wallet through a trade-only session key — trading rights, never withdrawal rights. It is built and testnet-verified, and the order-placement path is intentionally left unwired: nothing auto-trades. Shipping the capability and switching it on are two different decisions, and conflating them is how people get hurt.
The suite is not for sale as a consulting deliverable — the judgement that produced it is.