I connect ideas, people, and technology to turn complex problems into useful products. I’m a Computer Science undergraduate in New Delhi, with experience across enterprise workflows, cloud platforms, and tools for working with AI.
Portfolio ↗ · LinkedIn ↗ · Email ↗
Change the lens: The problem · The decisions · The evidence
Tap a project illustration to explore its logic.
Know what changed. Know what was checked.
A local-first behavioral diff for TypeScript and JavaScript. Connects code changes with their likely reach, verification plans, and recorded evidence through 17 MCP tools and a CLI.
Source · Published package · Case study
What counts as confidence after a code change?
A changed line is only the beginning. Veris connects it to a likely affected surface, a verification plan, and a record of the evidence. The distinction matters: a plan is a next step, and a record shows what was actually checked.
Security for what happens between agents.
A graph-based prototype for unsafe capabilities that emerge between individually authorised agents. Trace the finding, compare simulated containment options, and inspect the verification in a recorded engine replay.
Source · Replay console · Case study
Why look at the connections between agents?
Individually permitted capabilities can combine into an unsafe path. HIVE examines those relationships and makes the containment trade-offs inspectable. The public console is a recorded prototype replay, with simulated controls.
Explain why the argument persists.
An evidence-grounded account of disputed football decisions. IBM Granite reads the sources, Granite Guardian audits that reading, and explicit rules assemble the diagnosis. The public reader presents precomputed analyses.
Source · Evidence reader · Case study
Where does the model stop and the decision logic begin?
OFFSIDE separates source reading, an audit of that reading, and explicit diagnosis rules. That makes the route to an explanation easier to inspect. The public reader shows precomputed analyses; the engine lives in the source repository.
Follow another thread
- Pulse: monitor Claude Code sessions, steer tasks, and review tool approvals from a phone.
- Scrybe AI & Maniac: a public architecture portfolio for quantitative ML infrastructure and multi-agent systems built under BluOryn Technologies Pvt. Ltd.
- The full project directory: personal explorations, enterprise work, and hackathon builds, with the context behind each.
Curious by nature. Intentional by choice.
NIIT Learning Systems Ltd. · Project Intern · July–August 2026
Led OCR integration across requirements, architecture, workflow and field mapping, coordination, and testing toward production readiness. Contributed to an internal Network Intelligence Platform and performed AI chatbot UAT.
BluOryn Technologies Pvt. Ltd. · Platforms Management Intern · September 2025–May 2026
Led cloud platform modernization and worked across deployment, infrastructure, lead/CRM workflows, email and WhatsApp automation, and ongoing platform operations.
Proofr · Startup Catalyst · July 2025
Market research, TAM/SAM/SOM analysis, and competitive assessments for India’s early-career hiring ecosystem.
Foundation, community, and working toolkit
B.Tech, Computer Science and Engineering · MSIT, GGSIPU · 2024–2028
CGPA: 8.86/10. Currently Sponsorship Department Head at L.O.O.M, a student community built around learning, building, mentoring, and contributing.
Working toolkit
Python, Java, TypeScript/JavaScript, SQL, PostgreSQL, SQLite, MongoDB. AI-assisted work with Codex, Claude Code, and Cursor. Delivery with Azure Functions, Docker, Vercel, Neon, GitHub Actions, Dify, and n8n.
Counts come from the publicly visible GitHub contribution calendar. The chart prints its exact window and refresh time. Inspect the snapshot.
One small detour: play Signal Run
Sixty tiles. One red ball. Follow it through. Play Signal Run ↗
Keyboard and touch controls. This is a standalone game board, separate from the activity chart.
Stay interested after the first answer.
Start a conversation ↗