Muskan Kumari

About me

I am a researcher and developer focused on zero-knowledge proofs and zkVMs, privacy-preserving blockchain infrastructure, threshold cryptography and proof systems, and agentic AI — LLMs, multi-agent orchestration, MCP tooling, and AI security.

I am currently in my final year of my Bachelors at I.I.T Delhi. I am the President of the Blockchain Society at I.I.T Delhi.

I have worked as a Cryptography Research Engineer at Encifher, and as a Backend Engineer at Umbra Privacy.

Experience

• Umbra Privacy: Backend Engineer (December 2025 – April 2026)
  • Designed and implemented a privacy-preserving cross-chain bridge on NEAR Intents, used ephemeral keypairs and HKDF-derived keys to ensure source and destination transactions remain unlinkable on-chain, eliminating visible cross-chain actor graphs.
  • Built a production-grade Solana data indexer in Rust with Diesel ORM, handling complete historical backfills and real-time ingestion from RPC streams. Delivered consistent, queryable historical state for private transaction history and on-chain event reconstruction.
  • Solana Gasless Relayer: Developed a RESTful relayer service allowing users to pay fees in SPL tokens via Jupiter pricing integration. Implemented Redis-backed rate limiting and automated fee-payer signing.
  • Led end-to-end implementation of the Groth16 trusted setup (Phase 2) in Rust with snarkjs: Postgres-backed coordinator, Redis HMAC-signed job queue, Axum HTTP/WebSocket API, concurrent workers for powers-of-tau / R1CS / zkey processing, S3 artifact storage, timeout watchdogs, dead-letter queues, and API rate limiting. Dual-mode participant CLI (TUI and headless) for parallel multi-circuit contributions.
• Encifher: Cryptography Research Engineer (March 2025 – December 2025)

  Research and implementation on privacy-preserving key management and ZK for on-chain use.

  • Designed and implemented a Rust-based Key Management System (KMS) using threshold ElGamal encryption.
  • Implemented relaxed discrete-log DKG and Pedersen verifiable secret sharing for trustless key management.
  • Implemented a Bulletproofs-style protocol tailored for Solana, enabling efficient zero-knowledge range proofs.
  • Worked with elliptic-curve accumulators for privacy-preserving integrity checks and verifiable state consistency.
  • Led research and implementation across multiple encryption schemes, including ECIES and ElGamal.
  • Completed end-to-end STRIDE threat modeling for Encifher systems: mapped trust boundaries and data flows, identified spoofing/tampering/repudiation/information-disclosure/DoS/elevation-of-privilege risks, and documented mitigations to harden the KMS and cryptographic services.

Projects

• Compass — AI agent governance layer for Slack (2026–present): ASP.NET Core MCP and Agent Gateway that intercepts MCP JSON-RPC tool calls at the transport layer and enforces policy across multi-vendor agents (Claude, GPT, custom) without vendor code changes. Multi-agent security with confused-deputy detection, a 4-tier ring trust model, dynamic trust scoring, and per-agent circuit breakers; OWASP AIVSS guardrails with Microsoft AGT action taxonomy, prompt-injection detection, and ALLOW/DENY/PENDING decisions.
• Ooty VM (BrainfuckVM) — ETHIndia 2024 — STARK zkVM in Rust for Brainfuck: prover/verifier from scratch (custom FRI, univariate polynomials, Merkle commitments, Fiat–Shamir, finite-field crypto—no external crypto deps); univariate quotient polynomials and parallel Lagrange interpolation for faster proving vs. reference BrainSTARK
• Zk-Market — ETHGlobal Singapore 2024, Rootstock track winner — Privacy-preserving Web2/Web3 marketplace: proxy-based ZK-TLS and Reclaim to attest credentials and activity without raw data, trustless on-chain checks and incentives via Hardhat contracts
• proofPositive (Zcomp) — ETH Mumbai 2024, main-track winner in Zero Knowledge — Modular CirC-inspired compiler pipeline with frontends for C, Rust, Python, Solidity, and ZoKrates. Each path parses source to an AST, lowers through CIRC-IR existentially quantified circuits (EQC), applies backend optimization, emits R1CS, and synthesizes Groth16 proofs with bellman (Spartan also supported). The flow is end to end from high-level code through intermediate constraints to verifiable proofs, instead of authoring circuits by hand.
• Quark Layer — A decentralised proof verification and aggregation layer. Faster proof verification guarantees using cryptoeconomic security of restaked validators, called soft guarantee. A hard but slow proof verification guarantee using cryptographic security of zkVMs. Using aggregation to multi prove verifications and reduce on-chain verification costs. Reducing cost and latency by orders for proof verification(s).
• stark101-rs — Rust implementation of Stark101 by starkware.
• Rust Blockchain — Built a fully functional peer-to-peer blockchain in Rust implementing Proof-of-Work mining (SHA-256), libp2p networking (TCP + Noise, Floodsub pub/sub, mDNS discovery), block validation, longest-chain fork resolution, and multi-node synchronization. Developed a custom CLI supporting peer discovery (ls p), chain inspection (ls c), and block creation with network-wide propagation (create b).
• micrograd — Built a minimal autograd engine from scratch to understand how PyTorch computes gradients under the hood. Implemented backpropagation, computation graph visualization, and trained a tiny neural network using gradient descent.
• AlphaGraph — A multi-agent financial research system built with LangGraph that turns a stock ticker into a full investment memo with a Buy/Hold/Sell call. Seven specialized agents fetch financial statements (yfinance), compute profitability/growth/liquidity ratios in pure Python, and layer in news RAG (FAISS) plus LLM reasoning. Exposed as both a Streamlit app and an MCP server, with LangSmith tracing for observability.

Publications

• Encifher: A Trusted-Execution Coprocessor for Confidential Computation on Solana

Writings

• Ooty zkVM: Rust version of BrainSTARK — Part 1 — Introduction to zkVMs, ZKPs, virtual machines, and the Brainfuck ISA.
• Prerequisites for STARKs — Goldilocks field, roots of unity, Lagrange interpolation, Schwartz–Zippel, Merkle commitments, a full Fibonacci-square STARK walkthrough (LDE, constraints, composition, FRI), and Fiat–Shamir.
• PLONK-style arithmetization — PAIRs (preprocessed AIR), RAPs, multiset checks, permutation and evaluation arguments used in BrainSTARK-style designs.
• STARK Engine — How the Brainfuck zk-STARK engine fits together: VM ↔ prover ↔ verifier, register trace, and the processor, instruction, memory, and IO tables that the proof system reasons over.
• Arithmetization of Ooty VM — AIR constraints on processor, instruction, memory, and IO tables; permutation/evaluation arguments, terminals, and difference constraints for zero-knowledge.
• Stark Engine Overview — End-to-end zkVM workflow: trace → tables → codewords → quotients → nonlinear combination → FRI; ties to Ooty VM.

Academical Excellence

• ETHGlobal Singapore 2024: Zk-Market won the Rootstock track (best project).
• ETH Mumbai 2024: main-track winner in Zero Knowledge; best project among ~15,000 participants.
• JEE Advanced 2022: AIR 1147 (GE), AIR 8928 overall among qualified candidates.
• JEE Mains 2022: 98.3 percentile among 1.1M+ candidates.
• Cryptography Summer School (2025): certified participant — ZK proofs, SNARKs, MPC, FHE, blockchain.

Currently, I am interested on

• Zero-knowledge proof systems (STARKs/SNARKs) and zkVM engineering.
• Threshold cryptography and privacy-preserving infrastructure.
• Blockchain scalability, proof verification, and decentralised sequencing / DA themes.
• Mechanistic tools for reasoning about cryptographic protocols and compilers-to-circuits.