PPARVAAN / AHLUWALIA
Index/Project log/FIG. 05Applied AI

J.A.R.V.I.S. Assistant

A self-hosted AI assistant that lives in a single HTML file: local models first, voice in and out, and a memory that persists.

Built · iterating
OLLAMAVOICE I/O3D PARTICLES
LOCAL MODELS — OLLAMA FALLBACK — CLAUDE MEMORY BANK STL GENERATION VOICE I/O SYSTEM DIAGRAM
ProjectJ.A.R.V.I.S.
ViewSystem diagram
TypeSoftware
RunsLocal-first
Sheet05 / 06

01 The brief

Most AI assistants live on someone else’s server. J.A.R.V.I.S. runs on my own machine: language models run locally through Ollama, with Claude as a fallback when a task needs more.

02 What I did

  1. Built the whole assistant as a single HTML file, with a 3D particle-engine interface.
  2. Added voice input and voice output.
  3. Connected local models through Ollama: llama3.2 and qwen2.5:7b.
  4. Added Claude as a fallback model.
  5. Wired in live data feeds, STL generation and a persistent memory bank.

03 Engineering notes

Local first

Running models locally keeps conversations on my own machine and costs nothing per request. The fallback covers what the smaller local models can’t handle.

One file

Everything lives in a single HTML file, so it is easy to run, move and share, with no install step.

AI meets CAD

Because the assistant can generate STL files, it links the AI side of my work to the CAD and 3D-printing side.

04 Toolchain

InterfaceHTML · 3D particle engine
Local modelsOllama: llama3.2, qwen2.5:7b
FallbackClaude
FeaturesVoice I/O · live data · STL generation · memory

05 What’s next

→ NEXT

Coming next: a demo video of J.A.R.V.I.S. in action.