Hover the letters
I write the backend. The pipelines that move data around, the rules that decide who is allowed to see what, and the alerts that catch a problem before a customer does.
Four things I changed on a live AI teaching platform used by CBSE and ICSE schools. Each one started as something that was going wrong, and each one is measured.
Uploading a textbook meant pushing a whole 50–100 MB file through one request, and those requests fell over when the platform got busy. It now arrives as 3–5 MB chapter slices, in the background.
We used to learn a school was seeing errors days later — usually because the school told us. 19 alert rules now fire into Slack the moment latency or error rates move.
The same permission question was answered in 77 different places in the API, so any two of them could quietly drift apart. Every one now runs through a single deny-by-default path.
Switching a feature on for one school used to need an engineer and a code release. The business now flips any of 30 features, school by school, without us.
Hover a card — each one runs a small model of what it does
Scoring tenders · match fires an alert
An agentic Playwright scraper over Government e-Marketplace listings. An LLM scores every tender for relevance and bid-readiness, rows land in Google Sheets, and Telegram fires on a match — deleting a recurring manual review from a live procurement workflow.
Tracking people · nothing is stored
Real-time footfall counting, occupancy tracking and Gaussian heatmaps on commodity hardware — no GPU, no cloud, no stored imagery. YOLOv8n detection paired with SORT tracking behind a React and Vite dashboard.
Risk tiers · dashed line is intervention
Replaces manual review of student performance with automated risk tiering across 15 data models and 5 performance parameters. A ChromaDB and Groq RAG pipeline retrieves per-student context and drafts intervention plans for counsellors.
Filter the log
Software Engineering Intern · Bengaluru
Software Engineering Intern · Mumbai
Cut the list down
What I write in
4/6 in prod
Where most of my hours go
11/15 in prod
Models in a pipeline, not in a notebook
4/9 in prod
Enough to ship the whole thing myself
4/4 in prod
How the work actually gets shipped
7/7 in prod
The coursework underneath it
0/4 in prod
Backend, infrastructure or anything that needs to stay up. Fastest way to reach me is email — I answer every one.
email_address copied to clipboard
#eng-alerts · just now