Muskan Sharma — Main Researcher, Research Programmer, and Scientist at Data-T — leads FEATSRV and FeatLock 1.0: leakage-free point-in-time feature serving, open papers, and production-ready APIs for insurance MLOps.
Main Researcher · Research Programmer · Scientist · Data-T Quantum & Actuarial Research
Designs and ships end-to-end research systems — from temporal theory and synthetic benchmarks to FastAPI surfaces and operator consoles — so actuarial models train and serve on the same leakage-free feature definitions.
Complete programme for Muskan Sharma — research papers, technical blogging, and APIs — newest first.
Lead author programme on SCD Type 2 as-of joins, dual Feast/Dask and PySpark engines, Nimbus leakage traps, and FastAPI serving for underwriting, pricing, and fraud.
Muskan Sharma advances Data-T research through point-in-time stores, Redis online serving, leakage science, and open engineering reports.
Lead research platform for leakage-free actuarial feature serving.
Operator console and API surface for online, offline, and audit workflows.
Sub-10 ms policy and claim feature lookups for real-time scoring.
Zenodo archive and reproducible Nimbus fixtures for peer review.
Figure-led explainers on lineage, leakage, importance, and latency.
Same-origin /api/featlock proxy to FastAPI health, PIT, and audit routes.
Scientific and engineering highlights from Muskan Sharma as main researcher and research programmer.
Formalised valid-time as-of joins for SCD Type 2 actuarial entities so historical training sets never see future address, claim, or policy state.
Delivered Feast/Dask and PySpark SQL PIT paths with byte-identical regression checks for interactive and portfolio-scale workloads.
Shipped the operator-facing dashboard — overview, lineage, risk & leakage, feature importance, PIT training, and online Redis lookups.
Designed seven intentional leakage-trap packs so researchers can prove unguarded joins inflate metrics before production.
Authored FastAPI routes for /health, /online/*, /offline, /pit-training, /leakage-audit, and /feature-importance with x-api-key auth.
Bridges actuarial science and software engineering — reproducible notebooks, CLI service packs, and open-access Zenodo reporting.
FEATSRV engineering report documents PIT theory, dual engines, and Nimbus evaluation.
FeatLock 1.0 console visualises seven FEATSRV services for operators and auditors.
Online Redis lookups meet real-time underwriting and fraud scoring budgets.
Dask and PySpark PIT joins keep interactive and batch paths equivalent.
Seven Nimbus service packs stress-test leakage traps before model fit.
Papers, FeatLock APIs, and open-source research programming from Data-T.
What's New at Data-T — updates from Muskan Sharma's research programme, FeatLock releases, and papers.