Builder & creator
I make software for people. I find a problem worth solving, then build until it's solved.
I like to understand how things work, then build my own take on them. I wrote a programming language to learn how code really runs. I gave personal goals an AI product owner you can talk to on a live call. I built a guitar trainer that listens to you play, and a studio that turns an idea into a finished video. I work on every layer, from the first idea to the running product.
9
things built outside my day job
2
already running in the cloud
~90%
manual work removed by a GenAI platform I led at work
4
awards and hackathon wins
After hours
Companies have product owners.Now your goals do too.
Every goal gets an AI product owner with a name and a personality. It plans your week in sprints, checks in daily, picks up the phone for a live voice call, and reviews your week honestly until the goal is done.
Built with
Ideas in.Shorts out.
Start from a long video, a link, a topic, a script or today's trends. It researches, writes, voices, cuts and captions finished vertical videos, and can run the whole pipeline on a schedule with you reviewing before anything is published.
Built with
One skill. One screen.Focused practice.
A fretboard trainer for guitar players. It shows a note and a string, you find it on your real guitar, and the microphone checks your intonation. Levels, custom drills and a practice history keep you coming back.
Built with
Find musicians. Build a jam. Show up and play.
Musicians shouldn't need a WhatsApp group to play together. Jam handles discovery, forming a session, venue listings, group chat and setlists.
Built with
Drop in a pile of PDFs. Get clean data out.
Say what you want extracted and an LLM pipeline works through the documents, tracking where every answer came from. Swap between Bedrock, OpenAI and Gemini.
Built with
The right memory, at the right moment.
Capture lessons as you learn them. Before your next decision, ask, and it brings back what past you already figured out. Some memories can stay on your device.
Built with
Every office visit, from invite to check-out.
Staff invite a guest and each invitation gets a QR pass. Security pairs a phone as a scanner, checks visitors in at the gate, handles walk-ins and logs belongings, while admins see who came and when.
Built with
A GPT-style language model, built from the ground up.
I built a GPT-style model from scratch in PyTorch, with about 110M parameters, and trained it on Google Colab. Then I trained several specialised versions, including one that writes poems and one that writes like a particular author.
Built with
Specialised versions, including
A programming language of my own.
Rup is a small scripting language with its own interpreter, complete with classes, inheritance and closures. I then made it mine with built-in lists, maps and HTTP, a VS Code extension, and error messages with a sense of humour.
With thanks to Crafting Interpreters by Robert Nystrom, the wonderful book that taught me how languages work and guided me as I built this.
Built with
class Vehicle { move() { show "The vehicle is on the move!\n"; }}class Car inherits Vehicle { move() { super.move(); show "Vroom vroom!\n"; }}take myCar = Car();myCar.move();Output
The vehicle is on the move!
Vroom vroom!Craft
Mobile apps that feel at home on your phone, down to push notifications and live calls.
Complete products, from the interface through the API to the data model.
Agents, voice and retrieval that do real work for people, not just demos.
Batch and real-time pipelines that handle enormous volumes without fuss.
Owning what happens after launch: deployment, observability and cost.
At my day job
GenAI, at DISH
Finance work was stuck in slow, manual processes. I led the build of a platform that takes it on, with an LLM agent on an MCP architecture, multi-level approvals and real-time analytics.
React
Finance interface
FastAPI
Application services
LLM agent
MCP architecture
AWS Bedrock
Model access
PostgreSQL
Records, approvals
Cut manual processing by ~90% and earned the CFO AI Innovation Award.
Data platform, at DISH
Raw call records are no use to anyone as they arrive. One path turns high-volume call records into usage, call-quality and network-performance dashboards that teams can act on. A second path runs anomaly detection ML jobs on the data and raises alerts.
CDR sources
Diverse systems
Kafka
Event streaming
PySpark
Databricks, Dataiku
Iceberg
S3 data lake
Snowflake
Analytics
Elasticsearch
Kibana dashboards
Kafka
Event streaming
Python
Dataiku
Elasticsearch
Kibana dashboards
ML jobs
Anomaly detection
Alerts
Notify the team
Reusable components and automated validation improved data availability for analytics teams.
Research, IEEE 2023
An advanced warning mechanism, powered by renewable energy, designed to reduce conflict between people and animals on the road.
“Advanced warning mechanism to reduce man-animal conflict on roads using renewable energy,” IEEE, March 2023.
CFO AI Innovation Award
DISH India. For leading a GenAI platform that automated manual finance workflows.
WCA Innovation Award
For technology-led process improvements and rapid GenAI innovation.
Dish-way Champion Award
For technology-led process improvements and rapid GenAI innovation.
Winner, D'Spark Hackathon
For technology-led process improvements and rapid GenAI innovation.
Toolbox
Journey
Since Aug 2023
Data and AI engineering, DISH Network Technologies, Bangalore
Where I keep sharpening the craft: batch and real-time platforms on high-volume telecom data, plus LLM agents and MCP servers on AWS Bedrock.
2019 – 2023
B.Tech, Computer Science & Engineering, Jorhat Engineering College, Assam
Four years of computer science that ended with a published IEEE paper on keeping people and animals safe on the road.