Work

I founded Synapsed in 2022. It is a DPIIT recognised startup, and most of what it does is AI and business process automation for Indian companies, with lead generation at the centre of it.

Synapsed is where the client work lives. The jobs are mostly unglamorous and mostly the same shape: a business is doing something by hand that a system should be doing, and nobody has had time to fix it. So we build the extraction, the pipeline, the scoring, and the dashboard that finally tells the founder where the money is going. Business Connect ran us in their 20 Most Promising Startups list in 2023.

n8n, Next.js, Python

• synapsedinfotech.com
• Business Connect

Open source token reduction proxy for long running agents, built in five days during OpenAI Build Week and published to PyPI as agentwarden-ai.

Agents re-send the same context on every step and you pay for it every time. AgentWarden sits between the agent and the endpoint, spots the repetition, and strips it before the call goes out. The API key never leaves your process. It is the only thing here whose source anyone can read, which is why it is the first link I send.

Python, FastAPI

• Live
• PyPI
• GitHub

A drone swarm detection and threat assessment system. Finds every UAV in a feed, holds an ID on each through occlusion, and reads the swarm as a whole rather than as individual objects.

Tracking one drone is solved. Tracking twenty at once without swapping their identities is not. Once the IDs hold, the interesting part starts: convergence, formation and encirclement are patterns you can only see if you are looking at the group. It runs entirely on a laptop, no cloud and no API keys, because the environments this is meant for are the ones where connectivity goes first.

Python, FastAPI, ByteTrack, Next.js

Trading journal for retail derivatives traders. Organic growth, no ad spend, still running.

I built it because my own trading spreadsheet kept lying to me. Every user came from someone telling someone else in a Telegram group. Spending nothing on ads was not a strategy at the time, but it taught me more about how software actually spreads than anything I have deliberately tried since.

Next.js, Postgres

A smart money concepts indicator for TradingView. Sold direct with Razorpay and licence keys.

Building the indicator was a few weeks. Building the licensing, the payment flow and the support around it was everything else, and that ratio is the actual lesson.

Pine Script, Razorpay

A breakout system on the NIFTY SmallCap 100 and a carry system on GOLDPETAL futures, both running on Kite Connect and Dhan.

Paper results and live results are different animals. Everything I know about latency, partial fills and slippage I learned from the gap between the two, and most of that learning was expensive. They still run every session whether I am watching or not.

Python, broker APIs

Vectorised Python underneath all of the above.

Python, pandas

A lead generation site for a property developer, built around a real 3D model of the client's building rather than a photograph of it.

Compressing the model hard enough to open on a phone in Gujarat on a bad connection was most of the engineering. The model is the hook; everything beneath it is shaped to turn a visitor into an enquiry.

Next.js, three.js

Re-scoring dead pipeline and putting instrumentation on top of the CRM, so a business can see which stage is leaking revenue instead of guessing at it.

Most companies I work with are sitting on years of leads they wrote off, and the reasons they were written off are usually not in the system anywhere. So the work is two halves. First, going back through that history, re-scoring it against what actually closed, and putting the good ones back in front of a human. Second, flashing the data across the CRM so the founders can watch a deal move and see where it stalls. The second half is the one that changes how they run the company.

n8n, Python, WhatsApp API