Axioma AI

An enterprise AI assistant platform for the Axioma asset management system used in the energy domain.

Goal

The goal of the platform is to help users work with a complex enterprise system: understand domain data, navigate workflows, find the right information, and perform assisted creation or editing of business objects.

Architecture Overview

The platform is built as a private AI layer around enterprise data, domain documentation, and system APIs. It combines RAG, SQL access, norm-control workflows, local LLM infrastructure, observability, and LLM evaluation into a production-oriented assistant platform.

Axioma AI development and production architecture

Development and production layer: web UI, LangGraph orchestration, RAG pipelines, MCP/tool integrations, SQL connections, services, storage, and deployment structure.

Axioma AI local GPU and model infrastructure

GPU and model layer: LiteLLM gateway, local models through Ollama/vLLM, parsing/OCR components, and speech-related model services for private/on-premise AI usage.

Axioma AI monitoring and quality layer

Monitoring and quality layer: metrics, logs, dashboards, analytics, quality tracing, LLM-as-a-judge evaluation workflows, containers, alerts, and Langfuse-based LLM observability.

What I Built

  • Designed the architecture of a production-oriented AI platform rather than a simple chatbot.
  • Built multi-agent orchestration with specialized agents for different classes of tasks.
  • Implemented domain RAG pipelines for documentation, regulations, internal knowledge, and system-specific context.
  • Added a SQL sub-agent for structured data analysis and question answering over enterprise data.
  • Designed norm-control workflows to validate generated outputs against business and regulatory requirements.
  • Designed LLM-as-a-judge and automated evaluation workflows with Langfuse for agent quality control.
  • Integrated the assistant with enterprise APIs and internal services.
  • Deployed local LLM infrastructure on company servers for private, on-premise usage.
  • Set up Docker-based deployment workflows, CI/CD pipelines, and DevOps processes in coordination with system administrators.
  • Introduced a context-driven development workflow with architecture docs, global prompts, agent instructions, reusable skills, and project knowledge maintained alongside the codebase.

Stack

Python, FastAPI, LangGraph, LangChain, LlamaIndex, RAG, SQL agents, LiteLLM, Langfuse, LLM-as-a-judge, automated evaluation pipelines, PostgreSQL, Docker, CI/CD, local LLMs, Ollama, vLLM, REST APIs, Grafana, Prometheus, Loki.

My Role

Lead AI Architect / AI Platform Architect. I am responsible for architecture, backend implementation, agent design, local model infrastructure, deployment workflows, technical direction, and coordination of the AI development work.

I work with backend engineers, QA, product stakeholders, and domain experts while owning the AI architecture, core implementation, and technical direction of the assistant platform.