Embedded systems · Local AI · Agentic control · Asymmetric systems
Hardware + local inference + messaging interfaces that create private control planes over physical environments. Parallel focus: high-leverage agent infrastructure and distribution systems that compound into recurring revenue.
New Delhi, India · @karanjeet96
| Layer | Focus |
|---|---|
| Edge | Raspberry Pi, ESP32, ESP-CAM, sensors |
| Runtime | Local AI servers, private agent execution |
| Interface | API / WhatsApp / Telegram agent control, AI mic |
| Systems | Home Assistant, event-driven automation, isolation |
| Business | Managed OpenClaw / Hermes hosting, bulk LLM + VPS optimization, credit-capped subscriptions |
Raspberry Pi Home Assistant
On-prem smart-home control plane. Custom automations, sensor fusion, and decision loops that stay local. Zero dependency on external cloud for core reliability and latency.
Local AI Server
Self-hosted inference and agent runtime. Multi-model serving, continuous availability, and secured tool access. Backbone for private agent deployments and offline workflows.
Agentic Control via WhatsApp & Telegram
Natural-language agents exposed on the channels people already use. Stateful sessions → tool execution → physical or digital action. Permission boundaries and low-friction daily operation by design.
ESP32 AI Microphone
Edge voice capture and processing on ESP32. Low-power, always-available audio input that feeds local AI pipelines. Hardware + firmware path with no external service dependency for core features.
ESP-CAM + Sensor Workflow Automation
Vision + multi-sensor nodes. Event pipelines: detect → decide → act. Local processing for monitoring, triggering, and closed-loop automation in physical spaces. Deterministic behavior prioritized.
Agent Infrastructure
Validating managed hosting for OpenClaw (autonomous agents) and Hermes (self-improving agents). Core model: large VPS + optimized bulk LLM access → credit-capped user subscriptions. Exploring isolation, skill marketplaces, analytics, and white-label layers for recurring, high-margin revenue.
Distribution
YouTube systems at the intersection of AI workflows and Indian personal finance (Zerodha/Groww, tax, SIPs, FIRE). High-RPM evergreen content engineered as both audience acquisition and product validation engine (digital products, prompt systems, retainers).
Hardware → Agent Loop
Closing the gap so edge devices (ESP32 / Pi / sensors) become programmable through natural-language channels people already live in.
- Local-first and private by default.
- One system that serves many users or use-cases.
- Measure only what compounds: uptime, retention, unit economics, conversion.
- Reject mediocrity in architecture, isolation, and focus.
- Patterns and asymmetric upside over linear effort or hype.
X: @karanjeet96
Open to discussions on agent hosting economics, edge AI systems, and high-leverage product architecture.