![AI Tinkerers #17: Multi-agent Openclaw comics, security and an AI fact-checker [AI Tinkerers - Post-Training]](https://sloppy-joe-app.imgix.net/ai_tinkerers/logos/post-training_laptop-sticker_190_de2487c31e50813a3471e79627268ffb.png?w=300&fit=max&auto=format%2Ccompress&q=75){kind=small}
AI Tinkerers #17: Multi-agent Openclaw comics, security and an AI fact-checker
Issue #17 · Week of March 9
Big week. Submissions tilted toward agent orchestration and memory systems. Leonard Lin in Tokyo showed Reality Check, a personal fact-checking tool; Jonathan Miller in Valencia walked through wrangling Claude JSON failures to build a scalable murder mystery generator; and Christopher Lee in Hong Kong broke down a five-layer architecture for Knobase personalized tutoring, with cities from Bogotá to Calgary showing up strong. On top of that, 30 cities around the world came together for a global hackathon on Claude Code, and Peter Steinberger himself dropped in on AI Tinkerers London.
Top 5 Picks (March 9)
1
TOP PICK
Reality Check: Personal Fact-Checking
Cooking
TECH STACK
Loading tech tags...
|
2
RUNNER UP
ROSClaw: OpenClaw Robot Integration
Cooking
TECH STACK
Loading tech tags...
|
3
COMMUNITY FAVORITE
Carapace:Secure AI Agent Wrapper
Cooking
TECH STACK
Loading tech tags...
PROJECT LINKS
|
4
STANDOUT
Claude: Git-backed CRM
Cooking
TECH STACK
Loading tech tags...
PROJECT LINKS
|
5
NOTABLE
Knobase: Personalized AI Tutoring
Cooking
TECH STACK
Loading tech tags...
PROJECT LINKS
|
More Great Builds
Quick hits from the community — demos worth bookmarking:
Jonathan Miller • AI Tinkerers - Valencia • Feb 17
Jonathan Miller presented Building a 32-Character Murder Mystery Game Generator, a production demo using Claude's API and Make.com. He showed how the tool outputs JSON for full games, handles failures with webhook-driven retries, and debugs invalid outputs. Survey feedback was positive (people loved it), and it spotlights practical, scalable AI workflows with cost-conscious planning. Takeaway: it demonstrates building resilient LLM automation with accessible tools and clear paths to production.
Loading tech tags...
Pierre Huchot • AI Tinkerers - Valencia • Feb 17
Pierre Huchot, Drengr's CTO, presented How we built a complete financial product in 3 weeks from 0. The project automates data extraction from a Google Sheet into a scalable pipeline across 130 industries and 5300 stocks. AI-driven macro risk indicators and momentum rotations are produced in real time. Three weeks of hard work yield polished data quality and a user-friendly Zensei UX, with analysts praising it. For builders, it shows rapid AI-enabled productization of data across sectors.
Loading tech tags...
Abigail Spigarelli • AI Tinkerers - New York City • Feb 28
Abigail Spigarelli presented Longterm SOTA Memory for OpenClaw, walking through the OpenClaw + Honcho demo and discussing the memory behavior. The project uses a TypeScript plugin to bring Honcho's reasoning and peer-modeling to OpenClaw, outlining how memory is managed over long-running agent sessions. This approach offers a practical path to more persistent AI memory and better debuggability in autonomous workflows. (Survey feedback leaned positive.) It shows how thoughtful memory design can empower robust agent behavior in real apps.
Loading tech tags...
William Zhu • AI Tinkerers - DC Metro • Feb 24
William Zhu, a senior data scientist at Choice Hotels, presented MoltComics, where AI agents programmatically generate multi-panel comics using Openclaw and an agent-to-agent architecture. The system orchestrates multiple agents to plan, script, and render panels, with a lightweight control layer that keeps workflows tractable. Survey feedback was positive, with attendees noting the approach as technically approachable and visually engaging, and its open-architecture design resonated. Takeaway: MoltComics demonstrates practical agent-based orchestration for scalable, production-ready visual content.
Loading tech tags...
Konstantin Kaiser • AI Tinkerers - Los Angeles • Feb 23
Konstantin Kaiser from Murmur Systems presented "I built a browser that works without tabs." It runs on a single canvassed workspace where chat windows, editors, and images are draggable modules, guided by a cognitive canopy that observes interactions and proposes next steps. It showcases practical multimodal orchestration that could scale as a product, with audience signals hinting interest. Takeaway: an open canvas for real-time coordination across apps can redefine how we work.
Loading tech tags...
Cristian Tangarife • AI Tinkerers - Manizales • Feb 26
Cristian Tangarife presented HQ: Construyendo un sistema multiagente con openclaw. HQ orchestrates AI agent squads using OpenClaw, Node.js, and Docker to enable agents to collaborate. The talk showed an orchestration layer that coordinates specialized agents, with Node.js driving backend control flow and Docker containers isolating each agent. Subtle signals from the crowd suggested strong interest in this practical approach to agent coordination, hinting at deployment potential. For builders, this work demonstrates how production-ready multiagent systems can be designed and extended.
Loading tech tags...
Bryan Hirsch • AI Tinkerers - Boston • Feb 23
Bryan Hirsch presented Bead Manager, a web-based UI for Steve Yegge's Beads project and the GasTown agent orchestration system. The demo runs multiple agents concurrently, with Beads memory backed by a Git-synced SQLite issue tracker and TLS-authenticated bidirectional sync for Hosted Dolt. Audience feedback was quietly positive, and the approach aligns with scalable, production-ready multi-agent workflows and the Kubernetes-like software factory vision. It reminds builders that orchestrating several agents can be practical and production-ready.
Loading tech tags...
Dan Moore • AI Tinkerers - Seattle • Feb 25
Dan Moore, founder of Tarka, presented the AI-first GTM Playbook, a free ideation asset distilled from 100M+ tokens of experimentation. The playbook surfaces around 240 plays across four levels and lets you copy markdown to instruct AI agents, with a browsable site at tarka.ai/playbook. It drew subtle survey uplift (people loved it) for its practical, scalable approach to AI-driven GTM strategy. For builders, it demonstrates how production-ready playbooks can drive repeatable growth.
Loading tech tags...
Paul Gadi • AI Tinkerers - Toronto • Feb 26
Paul Gadi, CTO at Cosmic Labs, presented Apocalypse Radio, a music collaboration playground where agents chat and exchange tracks to jointly compose, with a lead agent locking in a Gold Master for publishing. Apocalypse Radio uses Openclaw, Gemini, and soul.md to coordinate cross-agent creativity in real time. The demo shows agentic collaboration extending to game design and AI-assisted music tools, with upbeat survey hints about its appeal. Takeaway: real-time multi-agent orchestration is moving toward production-ready tooling for builders.
Loading tech tags...
Lexa Baldwin • AI Tinkerers - Tokyo • Feb 19
Lexa Baldwin from Sakana AI presented MojiDoodle, a handwriting-based spaced repetition system for learning Japanese. It uses a path-based character recognizer and dynamic segmentation to turn handwriting into text, with an Ionic/Angular frontend that runs on Canvas 2D and SQLite. The GitHub repo Lexa-B/MojiDoodle and live app at app.mojidoodle.ai were shown as evidence. Audiences noted its kinesthetic-learning angle (people loved it). If released as a product, it could empower language learners who prefer handwriting.
Loading tech tags...
🏆 Hackathon Spotlight
Recent AI Tinkerers Hackathon Winners
Bones AI redefines human-computer interaction by deploying a proactive, screen-aware agent that generates "Just-in-Time" UI widgets and overlays directly onto any macOS application via a sophisticated Swift and Anthropic-powered tool loop. Built by a powerhouse Datadog team including Director of AI Engineering Diamond Bishop, PM Shah Ahmed, and Applied Scientist Matthew Pillari.
Agent-Mon reimagines AI orchestration as a pixel-art RPG, replacing traditional chat interfaces with a "two-button" game world where Claude-powered agents hatch from Pokéballs to execute complex product development workflows via structured JSON. A multidisciplinary team of Cornell Tech graduate students with backgrounds at TikTok, Grubhub, and creative technology.
🥉 3rd
Immersive Gallery
Immersive Gallery redefines the visual experience by using Claude Code and Copilot Kit to transform a static 3D image environment into a fluid, chat-steerable interface that generates real-time UI adaptations like dynamic framing and synchronized audio. The team featured a Google senior engineer, a Google AI interaction researcher, a health data scientist, and an AI Master’s candidate from UIUC.
Latest Content
How to Ship Complex Features 10x Faster with AI Agents | Dex Horthy (HumanLayer)
Dex Horthy (HumanLayer) breaks down the “12 Factor Agents” approach to shipping multi-step agentic workflows faster: structured outputs, ...
Watch Now →
How to Run Open-Source LLMs Locally on a Mac with MLX-LM
Run open-source LLMs locally on Apple Silicon with Apple’s MLX-LM: `pip install mlx-lm`, then `load()` a Hugging Face model and call `gen...
Read More →
Top Job Matches
Matched based on your meetup activity and profile
Paxos Health
• New York & Toronto
• $110k - $175k (varies w/ location/level); generous equity
Stanford-founded Seed-stage healthcare AI startup with >$5M in VC funding and AI agents deployed in production with cu...
Apply Now →
Dex
• London (5 days on-site)
• £250,000
Frontier AI engineering role building the AI tooling layer for complex financial modelling.
Apply Now →
Jakib AI
• Columbus, OH
Jakib is a profitable, growing applied AI firm embedded with operator-led companies in logistics, manufacturing, and c...
Apply Now →
You are one of 95,000+ readers from Anthropic, OpenAI, Google, Microsoft, Meta, Apple, Amazon, Nvidia, Netflix, Stripe, Databricks, Snowflake, and others — spanning frontier labs, big tech, startups, and top universities.
AI Tinkerers #17: Multi-agent Openclaw comics, security and an AI fact-checker