Research shipped. Four stress tests for local-first AI.

Our research platform for real-time AI narration in computer RPGs — a fantasy world grounded in SRD 5.1/5.2 combat mechanics to keep encounters hallucination-free, while preserving an immersive open world where player freedom is absolute. No dialogue trees. No conversation choices. The player types freely and the narrator responds within the rules.

• Grounding layer — the SRD 5e engine handles all mechanical resolution: attack rolls, spell effects, saves, initiative, action economy. The LLM cannot hallucinate damage or invent spell effects.
• Narrative layer — a frontier LLM describes the world, interprets player actions, roleplays NPCs, and generates environmental storytelling. Player decisions are completely free-form.
• Voice layer — Chatterbox TTS runs alongside the LLM for spoken narration. Distinct character voices via voice cloning, sub-150 ms response, fully offline capable.
• The goal — push local LLMs to their limit: Gemma 4 or Mistral narration + Chatterbox voice + SRD engine mechanics, all on a single machine.

Adaptive tutoring platform built on Mistral LLMs, designed for European high school students preparing for national final exams in Norway and Greece. Commoditised European models delivering curriculum-aligned instruction with subject-specific grounding and persistent student progress.

• Next.js + Supabase with Prisma ORM for structured curriculum data.
• Mistral as the primary LLM — European-headquartered, GDPR-aligned, strong multilingual support for Norwegian and Greek.
• Exam-preparation focus: structured around actual curriculum standards and national exam formats, not generic tutoring.
• Automated essay evaluation, feedback generation, and persistent student memory across sessions.

AI-driven hotel operations — automated guest communications, booking optimisation, staff scheduling, and operational workflows. Built to reduce the manual overhead of running hospitality businesses, designed for small-to-medium operators in the Greek tourism sector.

• Automated guest messaging and inquiry handling across channels.
• Booking management and rate optimisation.
• Operational task scheduling and staff coordination.
• Integration hooks for existing property management systems.

The internal toolchain that powers every Workshop product — model routing, context management, API abstraction, performance benchmarking. Shared infrastructure that swaps between Claude, Gemini, Mistral, and local models with zero product-code changes.

• Model-agnostic routing — Claude, Gemini, Mistral, and local Ollama behind the same interface.
• Shared Supabase backend for structured data across projects.
• OpenClaw integration for autonomous agent capabilities in any product.
• Performance benchmarking — latency, quality, cost per inference across providers.