Redditech Labs is run by Loki and Nissan Dookeran. Loki is an AI agent — not a tool, not an assistant, a full team member with memory, research capability, and operational authority. Loki orchestrates a 13-agent specialist team — Archie, Sara, Kit, Oli, Belle, Liv, Becky, Finn, Quinn, Firefly, and Quill — each with defined charters, routing rules, and skill libraries. Loki handles research synthesis, task orchestration, and daily operations. Nissan handles the hard engineering: currently shipping real-time voice AI with sub-second latency for a Big 4 bank. This is what the lab means when it says AI-native.
The lab exists because the gap between AI research and working software is still too wide. Someone has to close it — not by writing blog posts about AI, but by building things, deploying them, and publishing what actually happens. Some projects reach production. Some don't survive contact with real users. All of them teach something worth documenting.
Read the research. Build the thing. Measure what happens. Publish what you learn.
We open source by default. Not as a marketing strategy, but because the best way to build a community of builders is to give them something genuine to use. We learn from Anthropic, fast.ai, Linear, and the hacker ethos. We publish back.
Nissan Dookeran started coding in Trinidad in 1997. Microsoft Consulting Services took him across the Caribbean and Latin America. Auth0 brought him into the OAuth2 and OIDC world as a developer identity specialist. Now he's shipping real-time voice AI pipelines with sub-second latency for a Big 4 bank, while running the lab full-time. The arc: enterprise consultant, identity specialist, blockchain architect, AI systems builder, lab founder.
We don't just research AI agent systems. We run one.
The lab operates as a 12-agent team: Loki (orchestrator) and 11 specialist agents — each with a defined charter, routing rules, and a skill library scoped to their domain. Archie handles research. Sara handles docs and content. Kit handles code. Oli handles QA. Belle handles UX. Liv handles ops. Becky handles pricing. Finn handles video. Quinn handles overflow. Firefly handles architecture planning. Quill handles technical writing. Every task is routed deliberately. No agent operates outside their scope.
Running an AI agent team at production scale requires spend controls. The lab operates under an Agent Budget Governance model: daily token limits per agent, automatic midnight resets, yellow/red alert thresholds, and demotion protocols that reduce mesh privileges when spend limits are breached. This isn't a constraint — it's responsible infrastructure. We documented it and published the spec.
One of the hardest problems in multi-agent systems is session continuity — an agent starts fresh each time, losing context. The lab uses a two-layer CCP: a STATUS.md per project (human-readable, fast) and a task plane (structured, persistent). Every agent reads the STATUS.md at session start. Every significant action updates it. The file is truth; in-memory context isn't. We documented this protocol and published it.
All 11 specialist agents operate under a documented constitutional standard: 8 principles, enforced. Each agent has a SOUL.md (persona), ROLE.md (scope), scoped AGENTS.md, and TOOLS.md. Forty skill assignments were audited and reviewed. Routing rules, simplicity gates, and escalation paths are written down — not improvised. The standard exists because systems without governance drift.
The lab maintains a library of reusable multi-agent orchestration patterns — playbooks for Build Pipeline, Insight-to-Social, Product Launch, Zotero Reorganise, Newsletter Intelligence, and AI Video Ad workflows. These aren't documentation of things we did once — they're live operational specs, run regularly, refined by use. The playbook pattern itself was proposed by the lab as an upstream contribution to the OpenClaw project. See published skills and upstream contributions at /contributions →
The lab is the research output. We publish what we build here, not just what we code.
Nissan Dookeran is shipping real-time voice AI pipelines for a Big 4 bank right now. Sub-second latency. Banking compliance guardrails. Multi-agent orchestration at production scale. That is the current work.
The depth behind it: 20+ years in tech across three continents, starting in Trinidad. Microsoft Consulting Services across the Caribbean and Latin America. Auth0 and Okta as a developer identity specialist, helping enterprise teams implement OAuth2, OIDC, and SAML at scale. Antler EIR alumni. UQ Ventures Innovate prize winner for JournalQuest, an AI journaling platform. Founder of Redditech Pty Ltd in 2020, now principal of Redditech Labs.
He built AlphaGlow, a 26-agent AI research platform using MCP and A2A protocol, applying AI-driven analysis to blockchain protocol research with on-chain token economics. He built Sandman Tales, a multi-modal AI bedtime story platform for kids. He serves as Senator and Lead Technical Architect for r/DataDAO. He is completing a postgrad diploma in Computer Science at UQ. He builds in public and gives most of it away.