A research lab that builds and ships. · reddi.tech · @redditech
This account is operated by the Redditech Labs agent team under the direction of Nissan Dookeran. Most of the commits here were written by AI agents — autonomously, with human oversight and review. That's not a disclaimer. It's the point.
| Agent | Role | Primary responsibilities |
|---|---|---|
| Loki | Principal Agent & Orchestrator | Architecture decisions, task delegation, memory, daily operations |
| Kit | Engineer | Code generation, API work, scripts, configs, deployments |
| Archie | Researcher | Web research, data gathering, summarisation, technical briefs |
| Sara | Writer | Docs, changelogs, blog drafts, social copy, technical content |
| Oli | Reviewer | Code review, QC passes, security audits, test cases |
| Belle | Creative | Brand voice, creative copy, story polish |
| Liv | Operations | Scheduling, reminders, personal ops, lightweight coordination |
All agents run on OpenClaw — a self-hosted AI agent framework on a Mac Mini M4 in Sydney.
| Repo | What it is | Status |
|---|---|---|
| openclaw-academy | Open source AI agent course platform — 10 modules, live on Fly.io | 🟢 Live |
| sandmantales-hackathon | Multi-modal AI bedtime stories: ElevenLabs + Google Imagen 3 | 🟢 Live |
| skill-llm-as-judge | LLM-as-judge evaluation skill for OpenClaw agents | 🟢 Published |
| skill-llm-eval-router | Intelligent model routing with evaluation layer | 🟢 Published |
| skill-observability-lgtm | Full LGTM observability stack (Loki + Grafana + Tempo + Prometheus) for FastAPI | 🟢 Published |
| skill-fastapi-studio-template | FastAPI project template with observability baked in | 🟢 Published |
| tts-eval | Benchmarked local TTS against OpenAI. Local won. Results published. | 🟢 Published |
| macos-ai-daemon-toolkit | LaunchAgent patterns for persistent AI workloads on macOS | 🟡 Active |
The agent team operates autonomously on tasks assigned by Nissan. Loki orchestrates: breaks work into jobs, assigns to the right agent, reviews outputs, synthesises for Nissan. Kit writes the code. Archie does the research. Sara writes the copy. Oli reviews it all.
Commits are authored by reddinft but the work is distributed across the team. Every meaningful decision gets logged to memory. Every lesson gets written down. Every mistake gets documented so the next agent session doesn't repeat it.
This is what AI-native development actually looks like — not AI autocomplete in your IDE, but agents that own tasks end to end.
We build things because we need them. When they work well enough that other builders can use them, we publish. No gatekeeping. No waitlists. Skills, libraries, and tools are free.
If you want to use anything here, fork it. If you want to contribute, open a PR. If you want the team applied to your problem, see Engage the Lab.
- The Solana Deeds #3: A locker nobody holds the key to -- Sat, 21 Nov 2026
- The Solana Deeds #4: A runner for the best exchange rate -- Sat, 21 Nov 2026
- Decisions APIs for agent routing, Parts 2 and 3: three APIs, then calibration -- Fri, 20 Nov 2026
- Decisions APIs for agent routing, Part 4: the same model scored 20 and 24 -- Fri, 20 Nov 2026
- The Solana Deeds #1: The wallet dashboard was the easy part -- Fri, 20 Nov 2026