Aditya at a restaurant table, looking at his phone and smiling
that's me 👋 checking on my agents

Café owner · Trader · Self-taught builder

From a café counter
to AI agents.

I didn't come to software through a CS degree. I came through customers, markets, and a lot of curiosity. When AI made building possible for one person, I started shipping, and I haven't stopped. This page is that journey, with proof at every stop.

Open to roles and projects in AI agents & automation
4 yearsrunning a café, 100+ customers a day
1 live productVedPatrika: payments, WhatsApp, PDFs
24/7AI agents on my own server

The journey

Every stop taught me something the next one needed.

Always

The tech enthusiast who never shipped

I've always been the person who tries new tech first. But for years that meant using software, not making it. Building felt like something other people did.

2018 – 2022 · Gwalior

Kitty Café

I founded and ran Kitty Café in Gwalior. Within a year it was among the top 10 cafés in the city, serving 100+ customers a day with a team of eight. I handled complaints at the counter, vendors, inventory and the P&L myself.

COVID closed it. Hard lesson, and it pushed me to look at what else I could build.

What it taught meListen to the customer, then fix the real problem, not just the question they asked.

2022 – present

Markets

I started trading Indian equities, then moved into crypto and gold. Alongside, I took freelance technical support work: triaging API failures, server misconfigurations and webhook issues, and writing clear root-cause notes for clients.

Gold (XAU/USD) 15-minute chart on TradingView with entry, stop-loss and two take-profit levels marked
gold on the 15-min chart, levels marked
What it taught meRisk, discipline and journaling. Decisions from data, not from mood.
trading-mcp-server ↗ built later: NSE/BSE briefs + trade journal

2025

AI walks in

Last year AI changed how I work. With Claude and agent setups like OpenClaw, the gap between "I wish this existed" and working software got small enough for one person to cross in an evening.

So I stopped waiting to "learn to code first". I learned by shipping: small tools, then bigger ones, working through what the agents wrote and fixing what broke.

What changedCuriosity finally had a way out. The limit became taste and persistence, not a degree.

2026 · First builds

Tools for agents, then an engine

My first public projects were the things my own agents needed:

  • AgentRelay Apr
    WebSocket relay so agents can talk to each other with rate limits and an identity registry.
  • astro-skill Jul · open source
    Vedic astrology engine on Swiss Ephemeris: kundali, dasha, panchang, guna milan, bilingual PDFs. No API keys.
  • Hivelane Oct · launched
    Air traffic control for coding agents: routes each task to the agent that fits it, gives every worker its own git worktree, and hands the work on when a seat runs low.
hivelane ~ bash
Illustrative session. The commands and the route line are real, from the Hivelane page.

2026 · August

VedPatrika goes live

The astro engine became a real product at vedpatrika.in: a bilingual Janma Patrika anyone can order. I built and run all of it: the Next.js site, the Python API, Cashfree payments, delivery over the WhatsApp Cloud API with approved templates, email, free calculators and SEO.

VedPatrika homepage: 'Your birth chart — not just a chart, a clear reading of your life', with a green Janma Patrika book
vedpatrika.in: Hindi first, English on a tap
What it taught meShipping is the easy half. Payments, deliverability, support and Google are where products live or die.

2026 · Aug – Sep

A fleet that works while I sleep

I run a small team of AI agents on my own Linux server, controlled from Telegram: an assistant that ships code and runs ops, a content pipeline, scheduled research and reports, with shared long-term memory in Postgres.

Agent Notch on a Windows desktop: a slim pill of quota rings, glass hover cards with session and weekly usage, a refresh-quotas button and the settings panel
Agent Notch: every agent's quota at a glance

Now

What I'm building

Agent Notch: alerts the moment an agent is stuck waiting for approval, and one click to its terminal.
Hivelane: just launched. One agent in charge, the rest on call.
VedPatrika: growing it, one honest improvement at a time.

Proof of work

My GitHub from the last year. Most of my work is in private repos, so it shows up as squares, not names.

2,179contributions in the last year
123merged pull requests
11public repos

How I work

What the café, the markets and the code have in common.

Customer first

Years at a café counter. I find the real problem before I build.

Plain language

I explain technical things simply, in Hindi and English.

Ship, then verify

Small releases, checked in the real product before I call them done.

Honest numbers

No fake reviews, no inflated metrics. Proof over claims.