KS Khushvendra
Singh Resume

Systems engineer · IIT Jodhpur

I build systems that stay honest under pressure.

Distributed systems, open protocols, and applied AI infrastructure—designed for the messy edge cases where networks, incentives, and real users collide.

Explore selected systems GitHub
01 Summer of Bitcoin 2026 · ContextVM
02 Inter-IIT Gold Ranked 1st of 23 IITs
03 13 merged PRs Across 6 upstream codebases
04 IIT Jodhpur B.Tech · AI & DS · 8.20 CGPA

01 Selected systems

Work built from first principles.

Protocols and infrastructure where correctness, latency, and failure handling are part of the product—not implementation details.

01 · Protocol design / Active

CIPHER

Cryptographic Incentive Protocol for Honest Edge Routing

The problem

Edge relays are expected to deliver content honestly, but neither sender nor receiver should have to blindly trust the routing layer. CIPHER turns delivery into something that can be challenged, proven, and rewarded.

Core protocol

An O(1) challenge-response delivery proof built around a dual-nonce lottery, a Request Purgatory state machine, and Reed–Solomon recovery inside a QUIC-based Go daemon.

01

Constant-time verification per delivery chunk

02

Erasure coding for partial relay failure

03

Explicit state transitions for fair exchange

<10ms

proof overhead per chunk

<200ms

simulated delivery latency

O(1)

verification complexity

  • Go
  • QUIC
  • Cryptography
  • Game theory
  • Reed–Solomon
View source

02 · Open-source infrastructure / ContextVM

ContextVM

Decentralized MCP infrastructure, from transport to product.

Summer of Bitcoin · 2026

My contribution

One protocol stack. Three layers of work.

I contributed across ContextVM’s MCP-over-Nostr transport, trust-service infrastructure, and orchestration client—turning protocol behavior into reliable product paths instead of isolated features.

Transport

Deterministic relay discovery, server profiles, capability gating, and test hardening.

Service

Trust scoring over Nostr graphs, profile signals, and DuckDB-backed persistence.

Product

More dependable keys, model loading, tool-registry state, and agent orchestration.

01 / Protocol

SDK

MCP ↔ Nostr transports, relay discovery, signing, capability negotiation, and profiles.

Contributor fork 02 / Service

Relatr

A personalized web-of-trust engine exposed as an MCP service over decentralized identity.

Contributor fork 03 / Product

Web client

Server discovery, chat, payments, and agentic tool orchestration in a Svelte application.

Contributor fork
  • TypeScript
  • Nostr
  • MCP
  • Svelte 5
  • DuckDB
  • Bun
Explore ContextVM

03 · Distributed control plane

ARISTA RRM

Inter-IIT Tech Meet 14.0 · Gold

1 / 23 IITs

A distributed WiFi radio-resource management framework that turns noisy telemetry into stable, interference-aware channel decisions.

76%more stable configurations
40%QoE gain
≈0%neighbor interference
Key decision

Projected Gradient Descent for probabilistic graph coloring, backed by a high-throughput Kafka telemetry path.

  • Python
  • Kafka
  • MATLAB
  • ns-3
  • Optimization

04 · Peer-to-peer networking

Torrentium

Serverless file transfer, across hostile networks

A desktop P2P file-sharing system with deterministic discovery, parallel WebRTC transport, chunk verification, and peer reputation.

<500msRTT across peers
10+concurrent peers
30%reliability gain
Key decision

Kademlia DHT for discovery, WebRTC for parallel transport, and SHA-256 checks around every transfer boundary.

  • Go
  • libp2p
  • WebRTC
  • SQLite
Case study

05 · Mesh-native storage / Prototype

NostrMesh

An encrypted storage stack that keeps Nostr relay and Blossom services reachable across NAT through a private WireGuard mesh.

Architecture

Split metadata, content, and authorization into independent planes, then joined them with idempotent upload/resolve APIs, retry logic, and reproducible multi-node validation.

  • TypeScript
  • Nostr
  • WireGuard
  • Docker
  • PostgreSQL
  • Redis

06 · Applied AI / Cross-platform

AxeP

An attention exchange protocol joining native Android telemetry, fast WebSocket handshakes, and LLM-based state classification.

25%less battery use
>90%classification precision
  • Android
  • WebSockets
  • LLM

02 Experience

Working where protocols meet production.

Open-source agent infrastructure and distributed control systems.

May 2026 — Present
Summer of Bitcoin · ContextVM

Open Source Contributor

Contributing across ContextVM’s MCP-over-Nostr stack—from transport and relay discovery to service infrastructure and product orchestration.

  • Shipped CEP-23 server profiles and deterministic relay publication
  • Refactored oversized-transfer internals without changing the public API
  • Strengthened asynchronous transport tests and contributor documentation
ContextVM on GitHub
Sep 2025 — Dec 2025
Inter-IIT Tech Meet 14.0 · Arista Networks

Project Member · Gold Medalist

Built a distributed WiFi RRM framework for real-time channel allocation, combining a probabilistic optimization algorithm with streaming systems telemetry.

  • Ranked 1st among 23 IITs
  • 30% interference reduction
  • Up to 40% QoE improvement
Arista Networks

03 Open source

Contributions that made it upstream.

Selected merged work, organized by the systems it improved—not as a ledger of pull requests.

13verified merged PRs
6upstream codebases
4impact areas

01 Protocol & transport

4 merged

Predictable MCP over Nostr.

Made server discovery and metadata publication more dependable: shipped CEP-23 server profiles, deterministic relay behavior, modular oversized-transfer internals, and tests that no longer pass on unresolved asynchronous assertions.

Outcome

Cleaner transport boundaries and more interoperable discovery without expanding the public API.

ContextVM/sdk#43server profiles ContextVM/sdk#50relay discovery ContextVM/sdk#41transport internals ContextVM/sdk#34async tests

04 Field notes

Writing from the build log.

Practical notes on protocols, networking, and the decisions behind the systems.

Applied AI · Networks5 min

Your Router is Blind

How edge sensing, streaming telemetry, and guarded optimization make enterprise WiFi aware.

Read article
Open source · Protocol6 min

When AI Tools Need a Network

Discovery, signed metadata, trust, and product state across ContextVM’s MCP-over-Nostr stack.

Read article
Distributed storage6 min

Storage Has a Networking Problem

Inside an encrypted mesh that separates reachability, signed metadata, blobs, and authorization.

Read article
Browse all field notes

05 About

Curious about the machinery beneath the interface.

I’m Khushvendra, an undergraduate in Artificial Intelligence and Data Science at IIT Jodhpur.

I like problems that cross boundaries: networks with unreliable peers, protocols with incentives to model, and AI systems that still need predictable infrastructure underneath. That has taken me from MCP-over-Nostr transports to mesh-native encrypted storage—and from the algorithm all the way to production UX.

Outside code, I’ve led alumni initiatives that matched 25 mentors with 125+ students and helped secure ₹11+ lakh for campus coworking infrastructure.

Languages

Go · C/C++ · Python · TypeScript · SQL

Systems

Linux · Distributed systems · WebRTC · QUIC · Docker

Data & cloud

PostgreSQL · Redis · Kafka · GCP · AWS

Interests

Protocols · Fair exchange · Agent infrastructure · P2P

06 Contact

Have a hard systems problem, an open-source issue, or an interesting protocol idea?

Let’s build something that survives contact with reality.

khushvendras99@gmail.com
GitHub ↗ LinkedIn ↗ Resume ↗