Problems solved. Systems built.

Voice dictation and AI voice assistants are genuinely useful, but almost all of them work the same way: your audio leaves your device and goes to someone else's servers before you ever see a transcript. For people who talk about client information all day — lawyers, clinicians, accountants, advisors — that's not a hypothetical concern.

I built Kusema because voice tools usually send your audio to someone else's servers, and the people who'd benefit most (lawyers, clinicians, accountants, advisors) often talk about client information. Kusema transcribes speech locally, so no audio is uploaded. Only the text of a question goes to the AI provider you choose, or nothing leaves the machine if you use a local model.

It's a desktop app built with Tauri, Rust, TypeScript, and React, running local speech recognition (Parakeet, Whisper) and Kokoro text-to-speech on your own machine. Kusema is built on Handy by CJ Pais (MIT licensed) — it's an independent fork, not affiliated with or endorsed by Handy.

It's early software — version 0.x, currently for Windows 10/11 (x64).

Professional services firms share sensitive client documents constantly — tax returns, legal opinions, audit reports, financial statements, engagement letters. The default method is email attachments and shared drives. Once a file leaves the firm that way, control over it is gone entirely. It can be downloaded, forwarded, printed, and redistributed without the firm's knowledge. For firms with confidentiality obligations and compliance requirements, that's an unacceptable operational risk.

Sumanix is a secure document portal purpose-built for professional services firms. Documents are rendered server-side as secured images — the raw file never reaches the client's browser, device, or cache. Every rendered page is watermarked dynamically with the viewer's name, session identifier, IP address, and timestamp, burned directly into the image rather than overlaid on the original PDF. If a document is ever shared externally, it is traceable to the exact session that viewed it.

The platform includes full audit logging, exportable as CSV for compliance reviews and regulatory audits. Access is role-based — admins manage documents and users, clients see only their own files, with access granted or revoked instantly from any device. Word and Excel files upload directly and convert automatically to secured, watermarked documents with no manual PDF conversion required. Built-in redaction permanently removes sensitive information like SSNs or account numbers before a document is shared, burned into the file server-side so it can't be bypassed.

Sumanix runs as a hosted SaaS starting at $99/month, or as a self-hosted Virtual Private Cloud deployment for firms requiring complete data sovereignty over their own infrastructure. A newer Intelligence tier adds an AI assistant that answers questions across a firm's entire document library, with per-tenant data isolation and a guarantee that client documents are never used to train AI models.

East Orange is a city of 70,000 residents with no dedicated digital infrastructure connecting them to each other, to local businesses, or to city government in real time. City communications get lost in social media algorithms or buried on a website nobody visits daily. Local businesses have no direct channel to the neighbors most likely to walk through their door. Residents have no single place to find what's open, what's happening, or what the city needs them to know — right now, in East Orange specifically.

EastOrange.app is a free Progressive Web App — no app store required — built exclusively for East Orange. It operates across three interconnected layers.

For residents, it's a neighborhood platform with a live community board, local business directory, deal discovery, nearby map, event listings, safety alerts, and a loyalty punch card system for local businesses. At its center is an AI-powered neighborhood assistant trained on East Orange's streets, services, departments, and transit system — answering resident questions about city hall hours, which department handles specific issues, bus routes, and local businesses at any hour, instantly.

For local businesses, the platform provides direct visibility to East Orange residents specifically — not a general internet audience. A free listing covers the directory, AI recommendations, and the nearby map. A premium tier at $49/month adds featured placement in the deals strip, promotional posting, a verified badge, monthly analytics, and a built-in digital loyalty punch card program.

For city government, the platform pulls live content directly from the City of East Orange's official RSS feed, making city communications visible to app users automatically without any action required from the city. A municipal partnership enables targeted push notifications to all subscribers for emergency alerts, road closures, and community announcements — a direct line to residents who have opted in specifically because they care about East Orange.

The platform is built as a PWA — installable directly from the browser to the home screen, no app store, no download friction. No ads, no data sold, no algorithm pushing content from outside the community.

Inbound calls were going unanswered outside office hours. Leads called once and moved on to the next option on their list. Staff were spending significant time on repetitive intake questions that didn't require human judgment — time that could have been spent on billable work.

A 24/7 Voice AI receptionist handling inbound calls in real time, qualifying leads, answering common questions, and booking appointments directly into the calendar. Urgent situations are automatically escalated to a live person. The system handles the full intake conversation naturally without putting callers on hold or sending them to voicemail.

An organization managing sensitive case data for vulnerable populations needed AI capabilities for internal workflows but couldn't use standard cloud AI tools without violating confidentiality obligations. Their data could not travel to third party infrastructure under any circumstances.

A formal assessment and deployment proposal comparing two private AI paths — a managed private cloud deployment versus a fully on premise open source LLM installation. Each option was scoped specifically around the organization's data sensitivity, technical capacity, and budget constraints, with full cost, risk, and operational tradeoff analysis delivered for both.

Residents navigating permit and license applications were flooding municipal staff with repetitive questions — what forms to file, what fees apply, what the process looks like. Staff time was being consumed by inquiries that didn't require human expertise, pulling attention away from work that did.

A RAG based AI assistant prototype trained on municipal permit and license documentation. The system answers resident questions in plain language, surfaces the correct forms, and walks through process steps without staff involvement. Built as a proof of concept to demonstrate measurable impact before full deployment commitment.