About
I got here late, and on purpose.
Most of what I build sits between a language model and the person who has to depend on it. An inference server and the API layer in front of it. A platform for putting heavy models onto GPU machines. Retrieval pipelines, and the dashboards that make all of it operable by someone who would rather not think about GPUs.
I did not start here. My degree is in mechanical engineering, and I came to software properly through a full-stack course that finished the same month I started my first engineering job, in January 2024. Arriving late turns out to be useful: I still read documentation like someone expecting to be surprised by it.
The half of the job I actually enjoy is the unglamorous half. Naming things so the next person doesn't have to ask. Keeping state boring. Making a deploy repeatable with Docker and Nginx instead of memorable for the wrong reasons. When something costs me a day to work out, I write it up — that is where the Medium posts come from.
- Based in
- Noida, Uttar Pradesh, India
- Role
- Senior Software Engineer
- Writing
- 6 articles on Medium
01How I think about the work
The data model is the design
Most bugs I have chased were a schema apologising for a decision made in a hurry. I would rather spend an afternoon on the tables than a fortnight on the workarounds.
Retrieval beats a bigger model
When a RAG answer is wrong, the fix is almost never a different model. It is the chunking, the embeddings, or a search that never surfaced the passage holding the answer. The interesting work sits upstream of the model call.
Boring deploys
Docker, Nginx, a Linux box, and a process someone else can follow. Reproducible beats clever every time something breaks at an inconvenient hour.
Write it down
If a problem took a day to understand, the write-up takes an hour and saves the next person the day. That is the whole reason the Medium posts exist.
02What I build with
Languages
Python, JavaScript, SQL, HTML5, CSS3
AI engineering
LangChain, OpenAI SDK, Google Gemini, RAG, Vector search, Qdrant, n8n
Backend
FastAPI, Flask, Django, Node.js, Express.js, Kafka
Frontend
Angular, React, Redux Toolkit, React Router, Tailwind CSS, Bootstrap, Chart.js, Recharts
Data
PostgreSQL, MySQL, MongoDB, Redis
Platform
GCP, AWS, Azure, Docker, Nginx, Git, GitHub, Postman
The full list, including editors and the things I keep open all day, is on /uses.
03Where I have worked
- Mar 2026 — Present
Senior Software Engineer
Unicloud — Noida, India
Architecture, the API layer, and the management interface for the LLM serving stack.
- Designed the architecture and implemented the API layer and management interface for a scalable LLM inference server.
- Built a platform for deploying heavy-duty LLMs onto GPU virtual machines, supporting both serverless and clustered configurations.
- Improved build and deployment times by reworking how the Docker images were configured.
- Angular
- FastAPI
- PostgreSQL
- Redis
- Kafka
- Docker
- Jan 2024 — Mar 2026
Software Development Engineer
Tarini Consulting — Mohali, India
Full-stack product work on GCP and AWS, moving from conventional web applications into generative AI features as the tooling matured.
- Built web applications with React, Python (Flask), and Node.js on GCP and AWS.
- Integrated generative AI capabilities using Google's Gemini and the OpenAI SDKs.
- Developed and integrated RESTful APIs to retrieve and present dynamic data.
- Improved application performance and the experience on top of it by introducing a caching layer.
- React
- Flask
- Node.js
- Express.js
- PostgreSQL
- GCP
- AWS
Enough about me.
If any of this sounds like the shape of a problem you have, I would like to hear about it.