full-stack & systems engineer — graph-structured knowledge systems, terminal-native tooling, supply-chain security.
full-stack & systems engineer. Working out of India. B.E. · Chandigarh University · graduated June 2026.
use the tabs up top, the arrows, or your ← → keys.
I build systems where the structure is the point.
I build systems where the structure is the point — knowledge graphs with tens of thousands of nodes, dependency trees you can trace a compromise through, terminal tools that stay out of the way. Most of my work sits between a database and an interface, making large, tangled data navigable.
Currently a full-stack engineer at Chandigarh University's Lucknow campus, working on CMS architecture and a Neo4j knowledge graph over 20,000+ publications and patents. Before that I shipped a cross-platform desktop product at Notely on Electron and a microservices backend. On the side I write Rust CLIs and read a lot of lockfiles.
Backend, systems, and real-time work. Tap ⤢ to spread these out.
A terminal-native workflow engine. One pluggable abstraction over OpenAI, Gemini, Claude and Ollama with bring-your-own-key auto-detection, a multi-pane Ratatui interface, and an extensible command system for scripting AI tasks.
Maps npm and PyPI packages, versions and maintainers into a graph (HydraDB / OpenCypher), then answers the incident question: a package just turned malicious — which of my services are hit, how badly, and through what path. Blast-radius BFS, attack-path search, typosquat detection.
The backend for a desktop overlay assistant: it watches the active window and clipboard, collects and ranks on-screen context, resolves intent, and answers through OpenAI with a persistent memory layer — streamed over WebSockets.
Turns CCTV video into crowd-flow intelligence — density, direction, congestion and counter-flow anomalies — to flag crowd-crush conditions before they build, instead of just counting heads. Operator console over a computer-vision pipeline.
What a deploy platform does under the hood, rebuilt as three services — an upload service, a build/deploy worker, and a request handler that routes each request to the right build — wired with Redis, AWS and S3.
A Turborepo monorepo: authenticated real-time drawing over WebSockets with Postgres/Prisma persistence — front end, back end and shared packages in one graph, containerised with Docker.
A collaborative kanban board where every client stays in sync through a single WebSocket server broadcasting issue moves. Built on Bun, no database — the board state lives in memory.
Competitive programming, hackathons, and the odd trophy.
Notes on graph systems, terminal tooling, and backends. Full archive →
What changes when institutional research stops being rows in a table and starts being nodes and edges.
read →Building ChainTrace meant answering one question fast: if this dependency is bad, what breaks?
read →Terrek talks to OpenAI, Gemini, Claude and Ollama through one interface. The abstraction layer is the whole design.
read →SVG gave up around 800 nodes. Here's what moving the knowledge graph to canvas actually involved.
read →Building something that needs graphs, terminals, or a careful backend?
keizer · the laboratory · 2026
hand-bound with next.js. no template.