“Lifespan” Is Misleading
In CollectionScience has not extended the maximum human lifespan; it has raised life expectancy, which is a different thing. Why the ceiling near 120 years exists, read through an evolutionary lens.
Student. Computer Scientist. Chemist. Researcher. Builder of things at the intersection of quantum mechanics, generative AI, and synthetic biology.
Also: caffeine enthusiast, constructive empiricist, salsa dancer, and believer that the best ideas happen when disciplines collide.
I study Computer Science and Chemistry at the University of Toronto — two majors that keep dragging me toward the same handful of problems: quantum computing, machine learning, synthetic biology. This past summer I was a bioinformatics software developer at Roche's R&D division, building the computational infrastructure behind their new SBX nanopore sequencing platform. Before that, a year-long PEY co-op at Sanofi, writing high-throughput genomics pipelines and working out how to tell real viral mutations from sequencing noise. These days my research home is the Chemical Physics Theory Group at UofT, where I'm a research assistant in Professor Artur Izmaylov's lab. I work on sampling-based quantum diagonalization (SQD) built on Q-SENSE ansätze—the short version is that we exploit the structure of how electrons pair up to decide which slices of an astronomically large space actually matter for chemistry, which keeps the classical bookkeeping from exploding. It's the kind of problem where the theory and the numerics have to agree before you're allowed to believe either of them. (I also spent a couple of years on randomized quantum-simulation algorithms—qDRIFT, phase estimation, zero-noise extrapolation—but that line of work is on hold for now.)
I'm also a Teaching Assistant for theoretical computer science and calculus courses. I love teaching. Few things are as satisfying as watching students crack dynamic programming or finally understand why P vs NP actually matters.
But honestly? What gets me most excited is synthetic biology and AI. For the past three years, I've co-led the computational division of iGEM Toronto—the university's synthetic biology research team. We've won 4 Gold Medals and 2 Best Model Awards across four international competitions, building everything from phage engineering platforms to generative models for plasmid DNA.
I also write — there's a collection of essays below.
My current quantum work is with Professor Artur Izmaylov's Chemical Physics Theory Group, where I co-developed one of the first practical Sampling-based Quantum Diagonalization (SQD) pipelines built on Q-SENSE ansätze. The idea: seniority-structured states give you systematic control over which corners of Hilbert space actually carry chemical weight, so you spend your classical post-processing where it counts instead of everywhere at once. We designed numerical experiments tying that seniority control to sampling efficiency—and there's a particular satisfaction when the theory and the numbers finally shake hands. Manuscript in preparation.
Before that, I spent a couple of years on randomized quantum-simulation algorithms for electronic-structure Hamiltonians: product formulas (Trotter-Suzuki, qDRIFT) and block-encoding methods (LCU, qubitization). I built a framework pairing qDRIFT time evolution with quantum phase estimation—both QFT-based and Kitaev's iterative variants—plus a zero-noise extrapolation protocol using Chebyshev node sampling that kills the systematic O(t²) error and pushes toward Heisenberg-limited precision. That project is on hold for now, but the code is still up.
I like to think there is something fundamentally computational about biology; life is information...and Computer Science offers super rich frameworks to deal with information. Most of my projects on synthetic biology and bioinformatics—at least the ones not protected by confidentiality agreements—have either been crafted in conjunction with, or heavily inspired by, the iGEM Toronto synthetic biology research team.
I led and participated in the computational division for UofT's synthetic biology research team, competing in the International Genetically Engineered Machine (iGEM) competition. Four competitions. Four Gold Medals. Two Best Model Awards. Plus awards for Best Therapeutics, Best Entrepreneurship, and nominations for Best Foundational Advance, Best Wiki, and Best Presentation.
Built the first open-source platform for engineering bacteriophage receptor-binding proteins using generative AI. The challenge: phages are incredibly specific about which bacteria they infect, determined by tail fiber proteins that recognize bacterial surface receptors. Traditional phage therapy requires months of screening to find the right phage for a patient's infection. We wanted to design phages instead.
I integrated ESM3 (Meta's 98B-parameter protein language model) for sequence generation, Boltz-2 for structure prediction and binding affinity estimation, and MCMC optimization with simulated annealing to navigate the protein fitness landscape. Our energy function balanced predicted binding to target receptors, reduced binding to wild-type receptors, structural confidence (pTM scores), and sequence novelty. BLASTp confirmed all generated sequences had <40% identity to known proteins—we were genuinely exploring novel sequence space, not just interpolating.
We experimentally engineered four E. coli phages with retargeted host specificity: two to recognize truncated R1 lipopolysaccharide instead of K12 glycans, two to bind OmpC instead of LamB protein receptors. The platform reduces phage screening time from months to days. Wet lab validation ongoing; manuscript in preparation.
Developed a generative model for designing de novo plasmid sequences—the circular DNA vectors that are fundamental to synthetic biology but notoriously hard to design from scratch. We trained a Mamba2 state-space model (linear-time complexity beats transformers for long sequences) on 137,000+ plasmid sequences, with custom byte-pair-encoding tokenization and genomics-specific augmentations: reverse-complement and random circular crop strategies to respect the topological constraints of circular DNA.
In vitro experiments demonstrated that generated plasmids successfully replicated in bacterial systems, confirming our synthesized origins of replication actually worked. Plasmid.AI is now the largest open-source toolkit for plasmid foundation models, and we're continuing to improve it.
Applied constraint-based metabolic modeling and linear optimization (Flux Balance Analysis, FSEOF algorithm) to predict genetic modifications that enhance bacterial metabolic network efficiency. Think of it as computational strain design: given a genome-scale model of E. coli metabolism, which genes should you knock out or overexpress to maximize production of your target molecule? We evaluated over 12,000 genetic variations. Won Gold Medal + Best Model Award.
I've explored various corners of computational chemistry: ab initio methods like DFT and Hartree-Fock, molecular docking simulations, and unsupervised machine learning for drug discovery. Worked as a volunteer researcher at the startup Gene2Lead, applying clustering algorithms to different molecular representations for structure-activity relationship analysis. Also spent a memorable week debugging why HADDOCK3 kept silently dropping carbohydrate residues during protein-glycan docking (I wrote a 20-page troubleshooting document about it—the kind of painful experience that teaches you what computational biology actually involves).
Sanofi (May 2024-April 2025, PEY Co-op): Improved high-throughput genomics pipelines processing 4.5+ billion DNA reads per batch. Implemented alignment algorithms and Bayesian inference for viral minority variant detection. Developed novel heuristics using strand-bias correction and read-position mismatch analysis to distinguish biological mutations from RT-PCR and sequencing artifacts—reduced false positives by 15%. Created graph-based algorithms for detecting cross-sample contamination from barcode misassignment. Deployed optimized multi-threaded workflows on hybrid cloud clusters.
Roche (May-Sept 2025, Summer Internship): Developed computational infrastructure for evaluating Roche's novel SBX nanopore sequencing platform. Built bioinformatics pipelines for processing time-series performance data, enabling systematic benchmarking of read quality, throughput, and accuracy across experimental conditions. This work enabled the team to understand how algorithm improvements translated into real platform performance.
Most protein-ligand binding prediction methods treat the ligand as an afterthought—predicting sites from protein structure alone. LABind flips this with a ligand-aware approach that learns protein-ligand interactions simultaneously. We forked the original LABind architecture (Zhang et al., Nature Commun. 2025) and replaced the Ankh protein language model with Meta's ESM2-3B, upgrading from 1536-dim to 2560-dim per-residue embeddings. The hypothesis: ESM2's richer evolutionary patterns learned from larger-scale training might capture binding-relevant features that Ankh misses. Built collaboratively for CSC413 (Neural Networks) at UofT, this explores whether bigger protein language models actually translate to better binding site prediction—spoiler: sometimes yes, sometimes surprisingly no.
View on GitHub"Less of a structured research project and more of a frenzied, poetic excavation of quantum computation"—this repository explores amplitude estimation, linear combinations of unitaries (LCU), query complexity, quantum signal processing, and Hamiltonian simulation through the lens of both mathematical rigor and philosophical wonder. It's the computational equivalent of a mathematician doodling elliptic curves in lecture margins: structured notebooks on quantum algorithms interwoven with reflections on the nature of computation itself. Includes custom rendering functions for Hamiltonians in both explicit tensor product form and simplified algebraic notation. Named after Laplace's demon meets Schrödinger's cat, because quantum mechanics is fundamentally about what we can and cannot know.
View on GitHubSequencing platforms report quality scores (Phred scores) that are supposed to represent error probabilities: Q = -10 log₁₀(p). But how do you choose the parameters of a Beta(α, β) distribution such that the average error probability across n reads satisfies a target quality threshold with high confidence? This project tackles that question using concentration inequalities—Hoeffding's inequality for distribution-free bounds, Chebyshev's inequality when variance information is available, and practical quantile estimation for real-world constraints. It's the kind of statistical rigor that bioinformatics pipelines assume exists but rarely implement. Includes derivations, simulations, and reusable functions for modeling quality scores probabilistically rather than just trusting whatever the sequencer reports.
View on GitHubYou know that feeling when you encounter a real-world problem that perfectly maps to some obscure theoretical CS concept, and instead of using an existing library, you spend the next week implementing it from scratch because "it'll be fun"? Yeah, that's this repo. Reinventing wheels, diving into textbooks, crafting my own implementations. Sometimes the journey is the point.
View on GitHubYou can download my resume directly or view it in your browser:
Alongside research, I write essays about science — mostly about what is real, what is hype, and how to tell the two apart. The pieces below come from The First Second of Infinity, a collection on the biology of aging and the computational tools now orbiting it, developed over the past academic year with a professor as an independent studies course in writing and journalism.
A collection of essays on longevity science — and the noise around it.
Three parts. Part I treats the biology of aging on its own terms: what evolution selects for, the twelve hallmarks of aging, and how epigenetic clocks measure something we could not measure twenty years ago. Part II treats the marketplace built on that biology — the Sirtris saga, the venture capital, the supplements. Part III treats the two technologies whose promoters claim they are about to change everything: artificial intelligence and quantum computing. The throughline is skeptical: beneath most of the hype there is usually a real, slow scientific core worth watching.
Science has not extended the maximum human lifespan; it has raised life expectancy, which is a different thing. Why the ceiling near 120 years exists, read through an evolutionary lens.
The twelve hallmarks of aging as a diagnostic manual for mortality — twelve interacting failure modes, and why no single intervention closes more than one or two of them.
Epigenetic clocks turn invisible biochemistry into a single number for biological age — and the machine learning, from elastic-net regression to deep learning, that makes them tick.
A five-act story of resveratrol, Sirtris, and a $720 million mistake — how a flawed assay turned a measurement artifact into a pharmaceutical acquisition.
Cellular reprogramming and the Yamanaka factors — the science behind the first human trial of partial epigenetic reprogramming, and the scandals running alongside it.
What a quantum computer actually is, computationally — and a careful separation of what is real, what is noise, and what is probably wrong in the field.
"The merit of all things lies in their difficulty."
"Communication is one of those delightful things that only work in practice; in theory it's impossible."
"The metaphysicians of Tlön are not looking for truth or even an approximation to it: they are after a kind of amazement. They consider metaphysics a branch of fantastic literature. They know that a system (of thought) is but the subordination of all aspects of the universe to any one of them."But it hits different in Spanish:
"Los metafísicos de Tlön no buscan la verdad ni siquiera la verosimilitud: buscan el asombro. Juzgan que la metafísica es una rama de la literatura fantástica. Saben que un sistema no es otra cosa que la subordinación de todos los aspectos del universo a uno cualquiera de ellos."
'Zhuangzi and Huizi were strolling along the bridge over the Hao River. Zhuangzi said, “The minnows swim about so freely, follow- ing the openings wherever they take them. Such is the happiness of fish.” Huizi said, “You are not a fish, so whence do you know the happiness of fish?” Zhuangzi said, “You are not I, so whence do you know I don’t know the happiness of fish?” Huizi said, “I am not you, to be sure, so I don’t know what it is to be you. But by the same token, since you are certainly not a fish, my point about your inability to know the happiness of fish stands intact.” Zhuangzi said, “Let’s go back to the starting point. You said, ‘Whence do you know the happiness of fish?’ Since your question was premised on your knowing that I know it, I must have known it from right here, up above the Hao River.'
“It takes something more than intelligence to act intelligently.”
The moon has lost her memory. A washed-out smallpox cracks her face, Her hand twists a paper rose, That smells of dust and old Cologne, She is alone With all the old nocturnal smells That cross and cross across her brain. The reminiscence comes
“Every individual is at once the beneficiary and the victim of the linguistic tradition into which he has been born - the beneficiary inasmuch as language gives access to the accumulated records of other people's experience, the victim in so far as it confirms him in the belief that reduced awareness is the only awareness and as it bedevils his sense of reality, so that he is all too apt to take his concepts for data, his words for actual things.”
"To be shaken out of the ruts of ordinary perception, to be shown for a few timeless hours the outer and inner world, not as they appear to an animal obsessed with survival or to a human being obsessed with words and notions, but as they are apprehended, directly and unconditionally, by Mind at Large — this is an experience of inestimable value to everyone and especially to the intellectual."