Now · Dognosis
Dognosis's canine-olfaction program turns what dogs smell into data a model can learn from. I own the vertical that makes that data trustworthy — from the production backend down to the signal processing and statistics underneath.
Flagship · Sensor data → ML-ready datasets
The training-data pipeline resp → dogapi
I sole-architected resp (984 commits), the experimental pipeline where the hard parts were invented and stress-tested — GOR sniff detection, LSL↔Unix clock synchronization, three-way sample attribution, a logical-session resolver, and a versioned star-schema analytics layer — then hardened the proven logic into dogapi, the production FastAPI backend where I'm the top contributor and sole author of the dataset-generation and session-validation vertical. Together they fuse five independent evidence streams (IR array, wireless LSL, vision, QR, trainer-app state) into one confidence-tiered record per sniff, behind a physiological validation gate that keeps only trustworthy sessions in the training set.
Dognosis GitLab · dognosis/dogapi + dognosis/resp
The sensor systems it draws from
Olfaction platform
SniffSpace
A scent-detection platform where trained dogs evaluate samples while their physiology is recorded. I contribute to the multi-sensor acquisition pipeline — EEG, respiration, IMU and video — with Lab Streaming Layer synchronization, session logging, and data-quality verification across recording runs.
Canine brain–computer interface
DogSense
A wearable EEG system for dogs: a 3D-printed TPU helmet with a defined electrode montage, spring-loaded holders and an integrated amplifier. I work on signal acquisition and clock synchronization, and on machine-learning models that classify neural responses from the recorded EEG.
Selected Engineering Work
R&D · dataset pipeline
resp
The 984-commit R&D pipeline behind dogapi (sole author) — where GOR sniff detection, clock synchronization, three-way attribution and the logical-session resolver were invented, alongside a versioned MotherDuck star schema and an autonomous, leakage-aware ML research loop. Test code outweighs source code.
Quality · design-control system
otvcore
A git-backed, tamper-evident system of record for hardware design decisions — OTV / NCR / FMEA / ECO / SOP — as one engine with three clients (FastAPI + React UI, CLI, agent skill). It grew out of a cross-repo firmware audit and is built through an AI-agent workflow I design and direct, gated by a CI verify step.
Neuroimaging · MRI/CT pipeline & atlas
k9mri-tools
An end-to-end canine MRI/CT processing, audit and normalisation engine: converted 13 clinical studies (112 series) to BIDS with on-the-fly pydicom decoding for J2K-lossless CTs, audited voxel geometry via mirror symmetry (0 defects across 110 volumes), and achieved r = 0.862 / Dice = 0.887 nonlinear SyN registration to the Cornell 262-region canine atlas via a mask-confined objective.
Neuro · signal processing & ML
Canine EEG studies
An end-to-end EEG pipeline and studies on a g.tec rig. In an auditory-oddball ERP study I traced a ~1.45 s marker-timing bug that had nulled the effect, then proved the recovered response was neural — not cardiac — via R-peak subtraction and permutation testing. I also built a from-scratch ECG-on-EEG channel-viability detector, validated with nested leave-one-recording-out CV.
Diagnostics · data integrity
Data-integrity forensics
Root-caused silent data-corruption bugs in the training pipeline: crash-restart session contamination, a session that captured only ~57% of its duration, and a structural ~30% sample-code mismatch traced to a QR-scan vs TimescaleDB timing race — each with per-timestamp evidence and a precise query-window fix.
Research Path
Present
Data & ML Engineer
Dognosis
Top contributor to the production backend (dogapi): sole-authored the dataset-generation and session-validation vertical that fuses multi-sensor recordings into ML-ready training data. Designed and built otvcore (git-backed design control) and k9mri-tools (canine MRI/CT BIDS conversion, voxel-level geometry auditing, and SyN atlas normalisation). Also work across the canine EEG and olfaction stack (SniffSpace, DogSense): sensor acquisition, multi-stream synchronization, and machine-learning analysis of neural and behavioral data.
Oct 2022 – 2025
Postdoctoral Researcher
Decision Lab, IIT Kanpur
Studied the neural correlates of stress using a patch-foraging behavioral paradigm, combining EEG with behavioral analysis and convolutional neural networks to classify responses from EEG data. Awarded a two-year DST Cognitive Science Research Initiative (CSRI) Post-Doctoral Fellowship for a dynamical-systems account of patch-foraging behavior. Also contributed to the Rice–IITK collaboration on high-density, multi-region intracortical depth-array mapping during decision tasks — presented by invitation at Rice's InterfaceRice 2024 neuroengineering conference and published in NeuroImage (2026).
Nov 2019 – Sep 2022
Postdoctoral Fellow
Institute for Plasma Research, Gandhinagar · Basic Theory & Simulation Division
Modeled excitable media with cellular automata, showing how local timescales shape spatiotemporal dynamics — published in Chaos, Solitons & Fractals (2022).
2012 – 2019
Integrated MS–PhD, Physics
IISER Mohali · Supervisor: Prof. Sudeshna Sinha
Thesis: Dynamics on Spatially Extended Systems. Research on nonlinear dynamics, chaos, complex networks, and models of disease spread and persistence of infection.
2009 – 2012
BSc, Physics (First Class)
Loyola College, Chennai
Selected Publications
Optimizing electrode placement and information capacity for local field potentials in cortex
J. A. Willis, C. E. Wright, R. Zhu, Y. Ruan, J. Stallings, A. M. Abrego, T. Medani, P. Moitra, A. Ramakrishnan, C. E. Schroeder, A. A. Joshi, N. Tandon, R. M. Leahy, J. C. Mosher, J. P. Seymour · NeuroImage, 327, 121747
Impact of local timescales in a cellular automata model of excitable media
P. Moitra, A. Sen · Chaos, Solitons & Fractals, 162, 112418
Localized spatial distribution of disease phases yields long-term persistence of infection
P. Moitra, S. Sinha · Scientific Reports, 9, 20309
Emergence of extreme events in parametrically coupled chaotic oscillators
P. Moitra, S. Sinha · Chaos: An Interdisciplinary Journal of Nonlinear Science, 29, 023131
Anticipating persistent infection
P. Moitra, K. Jain, S. Sinha · Europhysics Letters, 121, 60001
Emergence of persistent infection due to heterogeneity
V. Agrawal, P. Moitra, S. Sinha · Scientific Reports, 7, 41582
Toolkit
Backend & Data
- Python · SQL
- FastAPI · Pydantic v2
- async SQLAlchemy · asyncpg
- PostgreSQL · TimescaleDB
- Docker · GitLab CI
- TypeScript · React
Signals & Neuro
- EEG · respiration · IMU processing
- MRI / CT processing & BIDS
- FSL · ANTs · AFNI · dipy
- Lab Streaming Layer
- Multi-sensor fusion & clock sync
- SciPy · MNE-Python
- Signal quality control
Modeling & ML
- Convolutional & sequence models
- Leakage-aware cross-validation
- Nonlinear dynamics & chaos
- Complex networks (NetworkX)
- scikit-learn · Pandas · NumPy
Beyond the Lab
I served as a reviewer for AIP Chaos (2022) and have moderated science-communication sessions at the India Science Festival, including a conversation on quantum computing with Ilyas Khan. Away from research, I've trained for 15+ years in Indian classical music, sung lead for a college rock band, debated at the national level, and captained a school football team.