Ranju GC

रञ्जू जि.सी.  ·  Research Engineer

I build systems that have to be trusted: threshold-signature bridges that move value between blockchains, and speech models for a language the internet mostly skipped.

Pokhara, Nepal ranjugc00@gmail.com github.com/gcranju Google Scholar

Three of five agree. The signature holds.

No single party holds the key. Take a share away and see what the bridge does.

नेपाली भाषामा बोल्न सक्ने प्रणाली Tacotron2 · 80 mel bins · 22.05 kHz · illustrative

Research

Two lines of work, one concern: building things people are asked to rely on, and being able to say why the reliance is warranted.

secp256k1 · ed25519

Applied cryptography and secure distributed systems

Multi-party computation and threshold signatures, cross-chain bridge and protocol design, and the integrity of decentralized systems once they are carrying real value.

WER · MOS · mel

Speech and language for low-resource languages

Text-to-speech for Nepali, dataset curation and public release, and evaluation of generative speech that goes past the loss curve to what a listener actually hears.

Publications

  1. 2024

    Optimizing Nepali PDF Extraction: A Comparative Study of Parser and OCR Technologies

    Paudel, P., Khadka, S., G.C., R., & Shah, R.

    Preprint · arXiv:2407.04577

    doi.org/10.48550/arXiv.2407.04577

  2. 2023

    Nepali Text-to-Speech Synthesis using Tacotron2 for Melspectrogram Generation

    Khadka, S., G.C., R., Paudel, P., Shah, R., & Joshi, B.

    SIGUL 2023 · ELRA/ISCA Special Interest Group on Under-resourced Languages, satellite workshop of Interspeech 2023, pp. 73–77. Peer-reviewed.

    doi.org/10.21437/SIGUL.2023-16

  3. 2022

    Decentralized Book Publishing Platform using Blockchain and IPFS: An NFT-based Solution to Royalty Theft

    Paudel, P., G.C., R., Shah, R., & Khadka, S.

    ICET 2022 · International Conference on Engineering & Technology

Systems

Showing

A Byzantine fault-tolerant bridge, built from first principles

A working simulation of a cross-chain bridge in which nothing is trusted but an honest majority of each validator set. Assets move between two simulated chains through three independent BFT sets, source, bridge and destination, and an untrusted relayer that carries the quorum certificate but whose honesty the protocol never rests on. Each set runs n = 4, f = 1, so the quorum of 2f+1 is exactly what the honest validators can reach on their own.

Every decision in both legs bottoms out in a single quorum function that enforces four things at once: signers distinct, signers authorized, keys bound to registered identities, and all signatures over the same canonical digest. A cryptographically valid signature over the wrong message therefore never counts, which is what most bridge failures reduce to. 105 transaction attempts cover every adversarial case I could pose to it, Byzantine signers, a tampering relayer, replayed certificates, repeated-signature stuffing, and 46 tests assert the invariant that matters: every token circulating on the destination chain stays backed one for one by value locked on the source.

PythonEd25519BFT consensusquorum certificates46 tests

A production cross-chain bridge, with no trusted relayer

Most people still move value between chains by trusting a centralized exchange to custody their assets along the way. This bridge removes that intermediary, which is exactly what makes its security assumptions load-bearing. An M-of-N verifier set attests independently to what happened on the source chain, and the signatures themselves come from NEAR Chain Signatures, NEAR's own decentralized threshold-MPC network. We used theirs; we did not build an MPC of our own. It signs across both secp256k1 and ed25519 chains.

Operating it taught me that the hard problems are not in the code so much as in the protocol: what exactly is trusted, what an adversary controlling some of the verifiers can do, and how value moves atomically across chains that cannot see each other. The quorum at the top of this page is that verifier set, in miniature.

M-of-N attestationthreshold MPCsecp256k1ed25519GoRust

Cross-chain intent relay

The cloud-deployed relay behind the bridge. It processes user transactions across heterogeneous chains, with modular per-chain deployments and infrastructure provisioned as code.

AWS LambdaSQS and SNSDynamoDBTerraformDocker

Sodax: cross-chain DeFi protocol architecture

Smart-contract architecture for a production hub-and-spoke money market and cross-chain swaps, with solver and executor services that route and settle user intents through the bridge above.

SolidityRustEVMNEARStellar

Multisig coordination tooling

A Go backend that coordinates multisig transaction proposals across chains, with wallet applications for Stellar and NEAR and a CLI for invoking contract methods on any of them. On-chain safety mechanisms including circuit breakers.

Gomultisigcontract auditing

Shruti: a Nepali audiobook reader

An end-to-end mobile application built on the Tacotron2 Nepali TTS system, carrying the research through to something a reader can actually listen to. The accompanying dataset is publicly released on OpenSLR. Training-data scarcity, and what it does to intelligibility, is characterized with word error rate, spectrogram inspection, and crowd-sourced mean opinion scores.

PyTorchTacotron2HiFi-GANrepository

Experience

  • 2022 — now

    Blockchain Engineer, Venture23

    Applied cryptography and secure cross-chain systems

    Bridge and verifier-set design, cross-chain relay infrastructure, multisig coordination services, and smart contracts in Solidity and Rust.

  • 2022 — 2023

    Undergraduate Researcher, CARD

    Center for Applied Research and Development, Institute of Engineering, Pulchowk Campus · Advisor: Dr. Basanta Joshi

    Worked on Nepali text-to-speech, a setting with very little prior public work. Curated and released a Nepali TTS dataset on OpenSLR, designed the evaluation protocol, and ran comparative experiments across TTS architectures.

  • 2021 — 2022

    Data Science Intern, Deep Analytics

    Mentor: Hari Khanal, PhD

    Air Quality Index prediction on real environmental sensor data for a Kathmandu-valley study, comparing and tuning Random Forest and Logistic Regression classifiers with attention to feature importance and temporal generalization.

Education

  • 2018 — 2023

    Bachelor of Computer Engineering

    Pulchowk Campus, Tribhuvan University, Lalitpur · 79.03%

    Excellence Scholarship on merit across all four years.

  • 2016 — 2018

    Higher Secondary, Science

    Amarsingh Model College, Pokhara · 3.71 / 4.00

Teaching

Mentoring

Mentor with Women Leaders in Technology since 2023, working with more than 50 fellows across three cohorts on AI and decentralized systems, and on how a technical career gets built. Technical sessions and workshops for student groups at PDSC, LOCUS, and STEM Sisters.

Community

Event Manager at Girls in Tech Nepal from 2020 to 2023, running more than 50 community events. PR Manager for the Hult Prize at IOE Pulchowk, reaching 12 colleges and 600 participants. Monthly technical-awareness sessions for secondary-school girls with Small Sisters, Pokhara.

Also

Awards

  • 2023Hackathon for Health & Tech, winner
  • 2023GritFeat AI Health Hackathon, 1st runner-up
  • 2022Hack-A-Week, winner, fintech
  • 2021Nepal AI School, full scholarship

Toolkit

  • LanguagesPython · Go · Rust · TS · Solidity
  • Machine learningPyTorch · scikit-learn
  • InfrastructureAWS · Terraform · Docker
  • DataPostgreSQL · MongoDB

Languages

  • Nepalinative
  • EnglishIELTS 7.5

R 8.0 · L 7.0 · S 7.0 · W 7.0