JUL 2025 → PRESENT
Quantum AI Researcher
Quantum optimization algorithms applied to large-scale combinatorial problems.
PHOENIX, AZ · QUANTUM · AEROSPACE · SOFTWARE
01 / ABOUT
In the span of a year I’ve worked on quantum optimization research, aerospace modeling and simulation, mechatronic systems, and a fintech product headed to production. Most people pick one of those lanes. The interesting problems live in the transfer between them: an orbital debris cloud is a combinatorial optimization problem; a flight simulator is a typed API with physics inside.
I ended up in quantum optimization because I believe it’s where the next generations of innovation will be defined. For a growing class of problems, classical computing is the ceiling and quantum is the way past it, which made the choice an obvious one: I’d rather learn to build there now than catch up later. The catch is that research doesn’t ship on its own, and bridging that gap is most of the real work. On Dextrivia, my orbital debris project, the quantum solver only means something because the classical pipeline came first: live orbital data, real delta-v costs, and a greedy baseline that any future quantum result has to beat.
The odd one out is LifeAndBooks, a book-and-life YouTube channel I run with my mom. Zero to 30,000+ subscribers, 215+ videos, 217K+ views: the editing, the branding, and a data-driven content strategy, all learned from scratch. It has nothing to do with qubits, which is exactly the point. The channel exists because I pay attention to the problems my peers and my generation are actually facing, and I want to put solutions, or at least honest conversations, within their reach, so the people around me can live more successfully, happily, and peacefully. It’s also how we carry the culture we come from and show it, proudly, to the outside world.
02 / EXPERIENCE
JUL 2025 → PRESENT
Quantum optimization algorithms applied to large-scale combinatorial problems.
JAN 2026 → PRESENT
Typed APIs, modern React frontends, and cloud deployment pipelines.
JAN 2026 → PRESENT
Modeling, simulation, and flight-systems tooling.
MAR 2026 → PRESENT
Integrating mechanical, electrical, and embedded software systems.
ONGOING
Mother-son book & life YouTube channel: 0 → 30K+ subscribers, 215+ videos, 217K+ views. Editing, branding, and a data-driven content strategy.
03 / PROJECTS
AI can write your code faster than you can understand it. Teams ship diffs nobody on the team can actually explain.
A git pre-commit gate that intercepts each commit and quizzes the author on their own diff (MCQ, matching, and free-response modes), blocking the commit below a 3.0/5 comprehension score.
1st place, NY Tech Week Intern Hackathon 2026 (Mantle, YC F25).
Removing debris from low Earth orbit is a sequencing problem: N objects, a delta-v cost between every pair, one optimal visiting order. Structurally, it's the Traveling Salesman Problem.
A pipeline that ingests live TLE data from Celestrak, propagates each object to a common epoch with SGP4, builds a Hohmann-transfer delta-v cost matrix, and solves a greedy baseline sequence for the Iridium-Cosmos 2009 debris cloud.
Establishes the classical benchmark a QUBO (quantum) solver is being built to beat. It's the project where the quantum and aerospace work meet.
Cloud code-review bots ship your diff to someone else's API and bury pull requests in low-value nitpicks.
A webhook server that pulls the unified diff on every PR event, chunks the hunks, reviews them with a local Ollama model, and posts inline comments anchored to file and line. When there's nothing worth saying, it deliberately says nothing.
A fully self-hosted review loop: no code ever leaves the machine.
Detective games replay the same scripted case; once you know the killer, the game is over.
An AI case generator (victim, suspects, motives, timeline, hidden killer) with a chat interrogation system where the killer lies and innocents don't, an evidence corkboard with contradiction tracking, and an AI-evaluated final accusation.
A complete noir experience: black-and-gold UI, typewriter effects, difficulty tiers, keyboard shortcuts.
Most RAG tooling assumes a cloud vector DB and a paid API before you can ask a folder of documents a single question.
A local-first CLI: `ragit index ./docs`, then `ragit chat ./docs`. Embeddings and generation both run through Ollama on your own machine.
Installable Python package with a working demo; zero external services.
Vocal feedback usually arrives after the performance: a coach reviewing a recording, hours or days later.
An ambient agent that listens live, analyzes pitch and delivery, and coaches in real time while you sing or speak.
Built end-to-end at a multimodal hackathon.
04 / STACK
FLIGHT STACK / PAYLOAD MANIFEST
05 / CONTACT
theatikshshukla@gmail.com