A local Generative AI assistant built to explore whether a self-hosted LLM could help with operational risk analysis and reporting — the kind of work that's normally done manually by reading incident logs and writing summaries for leadership.
🔗 Project showcase: https://vikrantthenge.github.io/ai-ops-assistant
A fully local AI pipeline — no cloud APIs, no external calls. Everything ran in Docker containers on my own machine:
- Ollama — running the LLM locally
- Open WebUI — Docker-hosted chat interface to query the model
- Prompt engineering — structured prompts for risk analysis, KPI assessment, and executive-style summarization, built around aviation safety and operational datasets
20+ years in airline operations means a lot of time spent reading incident reports, performance data, and audit notes, then turning that into a summary someone in leadership can actually use. I wanted to see how much of that first pass an LLM could realistically help with, running entirely on local hardware with no data leaving the device.
It worked — the model could take operational data and produce reasonable draft summaries and flag recurring patterns when prompted well. But running a local model alongside Docker and Open WebUI pushed past what my 8GB RAM laptop could handle at a usable speed, so I decommissioned the setup to free up the system. I don't have the original session logs or screenshots from that point — the showcase page describes the approach and architecture rather than reproducing a specific session.
Ollama · Docker · Open WebUI · Generative AI · Prompt Engineering
Built by Vikrant Thenge, drawing on 20+ years in airline operations across All Nippon Airways, Garuda Indonesia, and Uganda Airlines, with a current focus on operations analytics and decision-support systems.
