Optimal OffshoreProject 10 · AI tooling
VA AI Suite: an on-device AI console with WebLLM.
Suggestions, documents, invoices, a social scheduler and a prospect list — generated on-device with WebLLM, stored in Firestore, with no inference bill.
01 ▸ Overview
In-browser AI console.
VA AI Suite is an operator console for virtual-assistant work. Its defining decision is where the intelligence runs: a quantised Llama model downloads once, caches in the browser, and runs on the user's GPU through WebLLM — so there is no AI API, no per-token cost and no client data leaving the machine to be summarised.
Five modules ship live: caption, SEO and ICP suggestions in JSON mode; proposals, statements of work, onboarding documents and case studies exported as DOCX, PPTX or PDF; branded invoices with a draft → sent → paid lifecycle; a social scheduler with a daily publisher; and a prospect list with market-rate data.
Firebase is the record store and Vercel the host. A daily cron pushes due posts to a Make.com webhook for real publishing, and the app sets the cross-origin isolation headers WebGPU needs.
02 ▸ What we built
The parts that do the work.
On-device inference
WebLLM + WebGPU; the first run downloads the model, every run after is instant and free.
Document generation
Proposal, SOW, onboarding and case-study drafts rendered to DOCX, PPTX and PDF entirely client-side.
Invoices
Branded PDF invoices with status tracking in Firestore.
Social scheduler
Queued posts, a daily Vercel cron and a Make.com webhook for publishing.
Live operator dashboard
Prospects, records and outputs in one Firestore-backed view.
03 ▸ Stack
What it runs on.
04 ▸ Related services
The operations services behind the build.
05 ▸ Related projects