Sameer Bhalerao
All projects

Privacy-first life intelligence

Soul Spark

A multi-agent LLM advisor that unifies finance, health, career, and growth — all inference local via Ollama, no personal data ever leaves the device.

Live at soulspark.me

The problem

Personal data is fragmented across apps — health, money, tasks, behaviour all live separately — which makes it hard to turn into clear decisions. And the cloud-hosted LLM inference services that could help require routing sensitive personal data through an external provider.

How it works

  • Nine life domains — finance, health, career, social, growth, library, travel, inner self, and mindset — are unified into a single on-device DuckDB store.
  • A packet-native multi-agent runtime (an internal graph runtime with shared memory and a pending-action loop) reasons across domains.
  • All inference runs locally via Ollama — no personal data leaves the machine.
  • Real data sources include Apple Health and Google Sheets, ingested and unified into the on-device DuckDB store alongside finance, tasks, and behaviour signals.
  • A single conversational advisor moves from "here is your data" to "here is what you should do next" — the shift from dashboards to decisions.
  • Public access is via a Cloudflare tunnel; the data and the model stay on-device.

Architecture

Life domains

Finance

Health

Career

Social

Growth

Library

DuckDB

single local store, on-device

Multi-agent runtime

Internal graph runtimeShared memoryPending-action loopPacket-native contracts
local inference (Ollama) — no data leaves the device

Unified advisor chat

here is what you should do next

Privacy boundary

Everything above runs on-device. Public access only via a Cloudflare tunnel — no personal data ever leaves the machine.

Domains → one local store → multi-agent runtime over a local LLM → a unified advisor, entirely on-device.

The journey

Most personal-data tools show you one domain at a time — a health app here, a finance app there. Soul Spark unifies them on-device and turns the combined picture into a single, private advisor.

01

One daily read across your whole life — and an advisor that knows the context

Instead of opening five apps, you get one brief: what matters today across health, money, work, and the rest, written as guidance rather than raw numbers. Ask a follow-up and the advisor answers from your actual data — not generic advice — because it can see every domain at once. All of it runs locally; nothing leaves the device.

The daily brief — one read, framed as what to act on

Ask anything — answered from your real data, on-device

02

Patterns across domains — not metrics in isolation

The point of unifying the data is to see what no single app can: how falling sleep and rising workload combine into an energy drain, how a financial signal connects to a behavioural one. Soul Spark surfaces those cross-domain links, then collapses the breadth — nine domains, 70+ signals, 300+ connections — into one clear next action.

Connected patterns across domains, not isolated metrics

Nine domains, analysed together in one system

Highlights

  • 9 life domains
  • Apple Health + Google Sheets as live data sources
  • Local LLM (Ollama)
  • Multi-agent runtime
  • Fully on-device

Stack

PythonStreamlitOllamaDuckDBMulti-agent runtimeCloudflare

Why it matters

Explores decision-support over dashboards — and shows privacy-first design where it matters most: the moment AI touches your personal life.

Demo

Soul Spark — launch walkthrough