Skip to content
View stannum13's full-sized avatar

Block or report stannum13

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
stannum13/README.md

Shivank Nigam

Physics engineer building robotics systems, ML infrastructure, scientific software, and hardware-aware control.

I like projects where models and controllers have to work with limited measurements and messy hardware signals. My repositories include the code, experiments, failures, and current limitations—not only the polished demo.

Recent Robotics Work

A multi-month research program on the interfaces between semantic planning, persistent memory, learned policies, motion execution, and fast control. It contains reproducible MuJoCo and pure-Python studies, typed runtime contracts, and a 645-test suite.

An experimental harness for comparing deterministic and stochastic execution in ACT, VQ-BeT, and MolmoAct2. It measures repeated-output agreement, action drift, token entropy, inference latency, and deadline misses on fixed inputs.

A safety-first Unitree R1 simulation program with bounded cloud operations. The first completed infrastructure gate ran Isaac Sim 6.0.1 headlessly on an NVIDIA L4 and retained the reproducible environment, logs, cost, cleanup, and known shutdown issue.

Quantum Systems & Computing

Process-tensor world models for real superconducting-qubit data, spanning tensor networks, recurrent models, differentiable quantum channels, online learning, and delay-aware streaming inference. The project pairs promising results with matched Markov controls, transfer tests, and direct process-matrix physicality checks.

An interactive 3D architecture simulator for fault-tolerant quantum computing with 10,000 reconfigurable atomic qubits. It connects code construction and resource estimates to live controls for physical error, cycle time, runtime, feasibility, and qubit allocation.

A reproducible study of Fourier neural priors and causal noise forecasting for cyclic quantum-LDPC decoding with Stim and BP-LSD. Its fixed-shot experiments retain the negative results: learned priors hurt block-error rate, showing why syndrome validity and strong decoder controls matter more than imitation accuracy.

A matched-budget benchmark for real-time NV-center ODMR resonance and linewidth tracking. It includes causal playback, an eight-resonance virtual instrument, full-sweep and sparse estimators, and explicit accounting for acquisition, photon, timing, and compute budgets.

Primary Work

Measures a practical serving question: when many fine-tuned adapters share one base model, can requests be routed so the GPU reuses cached work instead of starting cold each time? Current work covers routing metrics, upstream pins, and smoke experiments; canonical SGLang results are not claimed yet.

Terminal tools for long tensor-network simulations, the kind that can run for hours before failing. It records structured logs, replays runs, and surfaces early warning diagnostics; canonical Quimb/TeNPy corpus validation remains pending.

Virtual lab for tuning photonic chips when heaters drift and every measurement costs time. It has controller baselines, drift/fault models, a SAX upstream pin, and a preregistered PhotonLock E001; the SAX adapter and canonical evidence are still future work.

Scores robot demonstration files before training, so bad or incomplete episodes can be filtered instead of silently poisoning a policy dataset. Current evidence is manifest and quality-feature smoke testing, not a policy-improvement result.

Focused Experiments

  • marginalia - tracks paper searches, quotes, citations, and evidence while building budget-aware research briefs.
  • policy-climb - deterministic benchmark for comparing small policy-search algorithms under the same evaluation budget.
  • linkscope - silicon-photonic link simulator that counts real Gray-coded bit errors through calibration and drift tests.
  • linebreak-uncertainty - BreakState tests whether small language models encode line position when text wraps to a new line.
  • memplex - Fisher Damping optimizer experiments for choosing natural-gradient damping from curvature signals.
  • qgf-autoresearch - QGF Ledger records bounded Q-Guided Flow replication attempts, artifacts, and failure modes.

Background

  • Dirac Labs - Founding Systems Engineer. Built sensor pipelines, Kalman filtering, and neural denoising for quantum magnetometry at 10-100 Hz real-time.
  • ETH Zurich - Thesis in theoretical molecular quantum dynamics. Accelerated nuclear-dynamics propagation using tensor-network methods.
  • EPFL - Thesis in quantum photonics. Simulated and fabricated diamond photonic integrated circuits for quantum sensing.
  • BITS Pilani - B.E. Manufacturing + M.Sc. Physics.

Contact

shivanknigam@gmail.com - LinkedIn

Pinned Loading

  1. qldpc-fno qldpc-fno Public

    Reproducible Fourier neural operator experiments for cyclic quantum LDPC decoding with Stim and BP-LSD

    Python

  2. oratomic-10k oratomic-10k Public

    Interactive 3D viewer and architecture simulator for fault-tolerant quantum computation with 10,000 reconfigurable atomic qubits

    TypeScript

  3. reflect-exp reflect-exp Public

    Reproducible robotics experiments on memory, recovery, action execution, and hierarchical embodied-agent interfaces.

    Python

  4. miniscale-pretrain miniscale-pretrain Public

    experimenting with pretraining models on publicly available data

    Python

  5. nonergodic-memory nonergodic-memory Public

    CPU-reproducible HMM-mixture belief geometry and causal-erasure research artifact

    Python

  6. wm-tn-worldmodels wm-tn-worldmodels Public

    Quantum process-tensor world models on real superconducting-qubit data, with matched controls, streaming benchmarks, and physicality checks

    Python