Member of Technical Staff & Co-Founder, ZeroShot
I build AI systems that hold up in production, from real-time speech and translation to secure cloud infrastructure, and help regulated teams adopt AI safely.
Sydney, Australia
Four areas where I do my best work, usually in combination.
Streaming speech recognition, live translation and captions that people can read while the conversation is still happening. I pay attention to latency, to how interim text settles into final captions, and to interfaces that work for people who are Deaf or hard of hearing.
Bringing AI into organisations where security, audit and compliance come first: clear boundaries on what data may reach a model, testing tools before they reach staff, and training that turns cautious teams into confident users. Much of my experience is in banking.
Java and Spring Boot services where correctness matters, such as payment and remittance messaging, approval flows behind one-time passwords, and encrypted document delivery. I also connect systems through the APIs and workflow tools they already expose.
Edge-first services on Cloudflare Workers and AWS, and the identity layer underneath them: Microsoft Entra ID, Intune, MFA and conditional access. I design services to hold as little data and as few credentials as they need.
Before choosing a model, decide what must never leave the device or the network. In Live Caption, audio goes straight from the browser to the speech provider and never touches our own servers.
Milestones are questions with a yes-or-no answer, not percentages someone has to estimate, so progress can be verified rather than argued.
Decisions are written down with their sources and confidence. When a claim turns out to be wrong, the record says why, so nobody repeats it.
Cost per audio hour, per request or per user is a design input from day one, not a surprise on the invoice.
Most automation comes from mapping the APIs and workflow tools a system already exposes. Custom code fills the gaps.
Follow the conversation. Take part in your own way.
A browser app that turns speech into readable, shared captions. A host starts captions in Chrome or Edge; others join by QR code or a six-digit code, read along on their own device and reply by typing. It is built for people who are Deaf or hard of hearing, for non-speaking participants, and for conversations across languages.
AwardPeople’s Choice Award, UNSW Founders Peter Farrell Cup 2026. Live Caption powered the live captions in WhoSays’ winning entry.
Microphone or tab audio is captured in the host’s browser.
Audio streams straight to the speech provider, and captions and translation stream back. Live Caption’s own services never handle audio.
Captions are encrypted in the browser and relayed to the room through Cloudflare Workers and Durable Objects.
Participants read along and reply by typing. The host can correct captions and export TXT, Markdown or SRT.
Sessions I run for teams adopting AI. Each takes about an hour and includes live demos.
How large language models work, what tokens and memory mean in practice, and the step from conversation to agents, together with the data-security rules for using AI at work.
Why the leverage has moved from prompts to context to the harness around a model: persistent memory, tool layers such as MCP, skills and hooks, state, and context compaction. Includes a method for designing reusable skills.
Jimmy Ji is a software engineer in Sydney and a co-founder of ZeroShot, an applied AI, data and engineering practice. He builds AI systems that hold up in production, from real-time speech and translation to secure cloud infrastructure, and helps teams in regulated industries adopt AI safely.
His background spans financial-services back-ends, enterprise identity and infrastructure, and AI enablement for staff. He created Live Caption, a browser app that turns speech into shared, translated captions for people who are Deaf or hard of hearing and for conversations across languages. He works in English, Mandarin and Cantonese.
Send the ZeroShot team a short note about the problem and where things stand. To reach me directly, email hello@jimmyji.dev.