Systems you own. Not software you rent.
I build self-hosted AI & automation on n8n and local LLMs — plus Python / TypeScript pipelines, agents, and scheduled jobs that run on infrastructure you control. Flat cost. Your data stays on your box.
IIT Kharagpur · 2026/self-hosted n8n + local LLMs/async, text-first/Kolkata
Tiny-Home Two-Way SMS Sales Agent
A quote page captures interest — then leads go cold before anyone replies, and a generic bot will happily invent a price it can't honor. So I built one that talks back, stays honest, and books the call.
- A web quote configurator feeds an n8n AI Agent running a real two-way SMS conversation.
- Window Buffer Memory holds context across the thread — it remembers what the lead already said.
- A local Ollama model grounded in a fixed FAQ, built to refuse to invent prices — it defers to a human instead of guessing.
- Lead scoring (HOT / WARM / COLD) decided in code, not the model — so qualification is auditable.
Every system, on the bench.
Real builds — n8n pipelines with local AI, Python & TypeScript automations, agents, and scheduled jobs. Each one runs on infrastructure you own. Hover a card to watch the signal move through it.
Rented software bills you for your own growth. I build the version you own — on a box you control, for a flat cost, with your data staying put.
— the whole point
Three things, each fixed-price.
- 01
Self-hosted automation, not metered SaaS
I rebuild your workflows on self-hosted n8n on a server you control. No per-task metering, no overage invoices when volume spikes — typically a flat, low monthly hosting cost instead of a bill that grows with your scale. You keep the workflows, the credentials, and full control.
- 02
Local LLMs + data sovereignty
I run open models (Ollama: qwen2.5-coder, llama3) on hardware you own, so lead, customer, and document data is processed on-premise and never sent to a public API. No per-token bill, no third-party retention — the right call when data is sensitive or per-call costs stop adding up at volume.
- 03
Two-way AI sales & support agents
Conversational agents that hold real two-way conversations — an n8n AI Agent with memory, grounded in your FAQ, built to refuse to invent prices, qualify each lead, and book the call. The same pattern routes support tickets by intent and urgency, with human approval gates where they matter.
I'm Soumyajit — IIT Kharagpur, class of 2026, building this from Kolkata. I skipped standard placements to build a backend AI & automation practice instead.
I'm early, and I take my first engagements at founder rates — so you get direct, founder-level attention and honest pricing while I grow my track record. I work async and write everything down: you get a documented architecture and a fixed number, not a calendar full of calls. Writing is where I'm sharpest — for this kind of work, that's a feature.
Tell me what's costing you.
Which metered tool is eating your margin (Zapier / Make / GoHighLevel), or what data you'd rather not send to a public API? I'll reply within 24 hours with a concrete architecture and a fixed price. Written is fastest.
async · text-first · no call required to begin