Open to interesting projects and conversations
That project became AtlasMind - an open-source Jira sprint visualiser built to solve a real problem Jira teams face every day: querying and visualising sprint data without writing a single line of JQL.
Ask questions in plain English, get back interactive hierarchy maps, tables, and AI-generated charts. It runs fully locally via Ollama — your Jira data never leaves your machine.
What it took to build:
- Local LLM orchestration with Ollama and Docker - reliable, reproducible, zero cloud cost
- Semantic search with pgvector for JQL annotation retrieval
- Self-hosted on Oracle Cloud with Caddy and Let's Encrypt
- The gap between "it works on my machine" and "it works for others" — the hardest lesson
AtlasMind reduced sprint query time from minutes to seconds — replacing manual JQL and EazyBi dashboard work for Jira teams.
→ Try AtlasMind · → View the code
llm-from-the-ground-up I wanted to really understand transformers — so I built a GPT-2 style LLM from scratch in PyTorch. Every layer, every matrix multiply, no black boxes.
NaamManthan A character-level generative model for Indian names, inspired by Karpathy's makemore. Small project, but one of my favourites — there's something delightful about a model that invents names.
docling-pgvector A RAG pipeline that turns documents into a searchable knowledge base using Docling and pgvector. The backbone of AtlasMind's JQL suggestion engine.
lida-ollama-extension · llmx-ollama-extension Extended Microsoft LIDA and llmx to support local Ollama models — because not everyone wants to send their data to the cloud.
If you're working on something in this space, I'm happy to collaborate:
- Jira & Atlassian tooling — custom AI interfaces, automation, natural language querying
- Local LLM integrations — Ollama, RAG pipelines, self-hosted AI systems
- AI-powered dashboards — replacing manual BI work (EazyBi, Tableau) with intelligent query layers
- Python & FastAPI backends — production-ready APIs with Docker and cloud deployment
I care about making AI actually usable — not as a demo, but as a tool that works reliably in the real world, on your own hardware, with your own data.
Local LLMs. RAG pipelines. Developer tooling. Natural language interfaces over structured systems.
Python · FastAPI · Ollama · pgvector · PostgreSQL · Docker · Oracle Cloud · Jira REST API · PyTorch · Caddy · LLM · RAG · Fine-tuning
📍 Berlin, Germany · 🌐 atlasmind.de
Always happy to talk about local AI, Jira tooling, developer tools, or anything you're building.