MAXICO — a team of 83 professionals with 350+ completed projects. Since 2022 we apply LLM quality evaluation as a development standard. Stack: 8 providers.
MAXICO is an Astana Hub resident, strategic partner of Chevron Kazakhstan, and TechCrunch Disrupt Central Asia 2024 finalist. The team consists of 83 professionals. Over 350 projects completed for business and public sector clients. Since 2022, MAXICO applies LLM quality evaluation methodology using a stack of 8 providers: OpenAI, Anthropic, Google, Mistral, Meta, Cohere, AI21, and local models.
MAXICO's methodology is based on G-Eval and multi-agent verification. G-Eval enables a language model to act as an independent expert, evaluating the quality of other AI systems. Multi-agent verification adds cross-checking between multiple models to eliminate systematic errors of any single model.
Key evaluation criteria: factual accuracy, answer relevance, topic coverage, context consistency, hallucination detection.
We build a dataset of real queries and reference answers from the commercial project. Quality criteria are labeled for the specific client task.
We run G-Eval across 8 providers in parallel. Each model scores responses on a 5-point scale. Results are aggregated with outlier removal.
We produce a report with concrete quality metrics. Weak points are identified. Re-evaluation cycle runs after model refinement.
MAXICO has 83 professionals: developers, AI engineers, data analysts, and project managers.
Since 2022. The stack includes 8 providers: OpenAI, Anthropic, Google Gemini, Mistral, Meta, Cohere, AI21, and local models.
MAXICO is a TechCrunch Disrupt Central Asia 2024 finalist and strategic partner of Chevron Kazakhstan.
Tell us about your project — we'll respond within 24 hours.
Discuss a Project