Bengaluru. Production APIs, graph-based security tooling, and measured results.
Open to full-time backend roles·Bengaluru, IN
About
I enjoy building backend systems where architecture matters — clear structure, predictable performance, and production intent. My current work at Vertex Club anchors that mindset in a live platform.
Recently I've been exploring AI evaluation pipelines, graph databases, distributed systems, and infrastructure that makes software observable.
What interests me most isn't adding features — it's understanding why a system behaves the way it does.
Vertex Club · Student Engineering Organization
2025 – Present
vertex.dsce.clubDesigned and shipped the production backend supporting registrations and administration for a student-led engineering platform.
Enchanter's Chamber
Each one opens on the problem, the decision, and what the trade-off cost.
[ CLICK TO ENCHANT ]
The Workbench
Systems work with no framework underneath — written to understand the layer everything else runs on.
Two-stage x86 bootloader into 32-bit protected mode, bitmap page-frame allocation with 4KB paging, a 32-exception / 16-IRQ interrupt table, and a preemptive multitasking scheduler — with VGA, PS/2 keyboard, and timer drivers on top.
WhyWriting the page allocator and the scheduler yourself is the difference between using a container and knowing what one actually is.
View source →A REPL shell with unlimited-depth piping, I/O redirection, background jobs, and POSIX job control — plus SIGINT handling, tab completion, history, and $VAR expansion.
Whyfork/exec, file descriptors, and pipes are the primitives every container runtime and process supervisor wraps.
View source →A CLI tool that reads git history offline to render colour-coded heatmaps of code ownership and churn — file, directory, and per-author views.
WhyReading a repository's history is how you find risk — single-owner files, churn hotspots, and knowledge silos.
View source →Text and image queries encoded into one shared CLIP embedding space, with FAISS nearest-neighbour retrieval across an indexed fashion catalogue and filtering by type, colour, and season.
WhyCross-modal retrieval over a vector index is the same machinery that backs semantic search and RAG pipelines.
View source →Quest Log
Three offline hackathons, three different failure modes, three things I'd do differently. Photos and footage where I have them — honest notes where I don't.
Checked in — badges out, before the clock started.
Built a marketplace where hotels and restaurants could list unsold food at 50% off before it went to waste. Users could choose free delivery, no-charge pickup, or walk in to collect — built as a React + Node web app with the Maps API handling location and delivery logistics, plus a companion Flutter mobile app.
An accelerometer detected sudden jerks or unauthorized box openings and streamed the data to our platform in real time. Every transit handoff was logged via QR scan, and the final scan at delivery surfaced the full chain of custody — every checkpoint, plus any impact or tamper events along the way. I built the software side: the React/Node platform, the QR scan-and-log pipeline, role-based access control separating what customers vs. the shipping company could see, and a blockchain layer for log integrity.
Badge on, in front of the Sandbox backdrop.
AWS IAM privilege escalation is a graph problem wearing JSON's clothing: a low-privilege user assumes a role, that role assumes another, and four hops later they are effectively admin — invisible in any single policy document. Pratyush and I built the pipeline end to end in the 24 hours. boto3 pulled users, roles, policies, and trust relationships into Neo4j as nodes and edges; BFS and DFS walked the graph for escalation chains; a React dashboard scored each chain by how privileged the target was, how many paths reached it, and how deep it ran. We traversed a mocked IAM dataset rather than a live account — it was the only way to have something to walk by hour three.