Background & Biography
His professional identity combines academic inquiry with responsibility for a company he helped establish. He describes an interest in mathematical problems that also have practical applications. That perspective brings together cryptographic methods, the rules by which distributed participants reach agreement, and the incentives shaping their behavior.
His assistant professorship sits within a computing department whose directory places him among its cryptography faculty. The listing connects his research to blockchains and techniques for demonstrating knowledge without exposing the underlying information. His institutional office establishes a professional association with the country recorded here; it does not establish personal residence or citizenship.
Notable Contributions
The 2018 FlyClient presentation, coauthored with Mahdi Zamani and Loi Luu, tackles a basic obstacle for lightweight clients: deciding which chain to accept without downloading every header. A prover supplies evidence for randomly selected blocks, while a Merkle mountain range ties those blocks to positions in the same history. This prevents a purported proof from simply assembling unrelated, correctly mined blocks. The client can then examine transaction inclusion against that chain evidence. The presentation describes a prototype, with guarantees tied to its sampling and adversary model. Checking a sampled history should not be confused with independently validating all state and transactions.
His coauthored work on batching set accumulators addresses a different storage problem: proving that an output remains available to spend. A compact commitment represents the relevant set, and an accompanying witness supports a claim of inclusion or exclusion. Combining many such checks can reduce repeated verification work. In the proposed blockchain application, users maintain their outputs and the evidence needed to spend them, while validators can work from a smaller representation of the set. Additions and deletions still require correct updates. The contribution changes the distribution of storage and checking responsibilities; it does not make the underlying records or witness maintenance unnecessary.
In the DARK research with Alan Szepieniec and Ben Fisch, the question shifts to the assumptions behind short proofs. The associated presentation constructs commitments to polynomials using groups of unknown order, with class groups offering a way to avoid a secret setup ceremony. These commitments can turn a suitable interactive proof into an argument that is compact to transmit and check. That route changes the source of trust, rather than abolishing assumptions. The construction still depends on mathematical hardness and the model used to replace interaction with hashing. Its significance lies in making setup requirements an explicit part of proof-system design.
The coauthored work Proof-Carrying Data without Succinct Arguments asks how to keep proving a computation as it grows. Rebuilding one large proof each time another step is added can repeat work and increase memory demands. The presentation develops split accumulation: relatively small checks accompany successive updates, with a separate final decision procedure checking the accumulated evidence. Crucially, the construction can use underlying arguments whose proofs are not themselves short. Its concrete route relies on discrete-logarithm assumptions and a random-oracle model; recursive composition also carries a stated heuristic qualification. This is accumulation of proof obligations, distinct from maintaining membership in a set.
The coauthored presentation Proofs for Inner Pairing Products and Applications studies how to compress checks involving many cryptographic pairings. Its presentation includes an application that combines compatible proofs into one shorter object for a verifier to inspect. The aggregator performs work that would otherwise be repeated across separate checks, while the verifier still processes the relevant public inputs. The slides distinguish constructions using specially structured setup material from alternatives with different verification costs. They also report a research implementation. These are particular algebraic techniques with defined prerequisites, rather than a claim that any collection of proofs can be compressed in the same way.
The HyperPlonk work, coauthored with Binyi Chen, Dan Boneh and Zhenfei Zhang, concentrates on the cost of producing a proof. Its presentation represents a computation using multilinear polynomials over a Boolean hypercube and uses sum-check machinery, avoiding the Fourier-transform step emphasized in its comparison. It also supports more elaborate custom gates, so some application operations can be expressed without decomposing them into as many small pieces. Lookup techniques extend that approach to checking values against prepared tables. The slides discuss implementation results and remaining bottlenecks. Actual costs and security conditions depend on the selected commitment system and the computation being proved.
A 2023 paper with Sourav Das and four other coauthors addresses signing groups whose members carry unequal weights. It supports multiple approval thresholds without treating a heavily weighted participant as numerous separate signers. An inner-product argument helps establish that the combined signature meets the required weight. Compact signatures and verification material are balanced against substantial preparation by the aggregator, including preparation effort quadratic in the size of the signing group. The construction requires a powers-of-tau setup and registered public keys, with security analyzed in the Algebraic Group Model. The authors also implemented a smart-contract verifier. The resulting signatures lack uniqueness, limiting their use for producing shared randomness.
Why Katharos.ai Follows Benedikt Bünz
Katharos.ai follows this research because proof generation is a concrete infrastructure constraint. Changing how a computation is represented can affect who can produce its evidence and what hardware resources the task demands.
Collective authorization is another relevant thread. A system may need to verify sufficient support without assuming that every participant has equal influence. The weighted construction makes that distinction technically explicit and exposes the costs of supporting it.
Areas to Follow
- Lightweight verification: whether a proposed use preserves the sampling conditions and adversary assumptions behind the evidence, and which transaction or state checks remain outside it.
- State maintenance: who updates spending witnesses after the committed set changes, and whether batching reduces overall work or shifts more responsibility onto another participant.
- Continuing computation: how an implementation separates recurring checks from the final decision, and whether its composition argument preserves the assumptions of its underlying components.
- Proof aggregation: which incoming proofs are compatible, what setup material the construction needs, and how public-input processing affects the verifier as the workload grows.
Reporting Scope
This profile covers documented professional roles and selected collaborative research. The technical discussion draws on six presentations and one conference paper; the presentations are used as presentations, not represented as complete papers. Reporting distinguishes a proposed construction, its security model and a research implementation from evidence of operational adoption. It makes no claim that a cited design is currently deployed by his company, or that dated experiments establish performance for a present-day service.
Areas of Influence
His work is relevant to decisions about where a cryptographic system places trust. A short proof can be attractive while its setup remains consequential. The transparent-construction research gives developers a concrete way to examine that tradeoff alongside the mathematical assumptions, communication requirements and verification work it entails.
Photo: NYU
Katharos.ai publishes this profile for information, based on the sources listed. Inclusion is not an endorsement of the individual, their organizations, or any investment product.
Source notice: Information and images are attributed to the external sources listed. Katharos.ai does not endorse the subjects or claims described and does not independently certify the authenticity of every source statement or image.