VAIBHAV KHURANA / AI + DATA SYSTEMS
Data and AI systems that hold up.
I build reliable data platforms, applied AI, and analytical products from source material through deployment.
LIVE PROJECTS
A working portfolio, not a highlight reel.
01Legal Discovery Intelligence GraphA public synthetic-matter investigation workflow combines vector and graph retrieval, cited evidence, timelines, privilege/PII flags, calibrated refusal, and an auditable case brief.02Legal Document Intelligence RAGA versioned public-document pipeline turns court opinions and SEC filings into 3,055 searchable chunks and returns evidence-linked answers with an explicit refusal path.03Text-to-SQL Interface with Guardrails and Hallucination DetectionReviewers can inspect SQL, assumptions, schema lineage, policy checks, and a bounded result preview before one single-use execution; unsafe, malformed, identifier-exposing, and hallucinated-schema queries fail closed.04Automobile-Loan First-EMI Default Strategy PortfolioA credit-policy analyst can test where to draw a first-EMI risk line, see the confidence interval around the answer and who it declines, and take a recommendation or a refusal to governance. On the published assumptions the honest output is a refusal: no evaluated band clears zero.05Credit Risk Model Validation & Review-Capacity LabThe selected model clears prevalence/random, repayment-delay, and logistic references on repeated paired development evidence, but ties calibrated Extra Trees because the PR-AUC advantage does not clear the prespecified practical margin. At 10% holdout capacity, 600 historical rows contain 431 observed defaults with 71.8% precision and 3.25× lift, each reported with uncertainty.06Application Fraud Strategy PortfolioA fraud strategy analyst can compare screening approaches at a fixed review capacity, see what each buys and costs, and take a recommendation or a refusal to governance. On the pre-agreed checks the honest output is a refusal: the proposed model catches 472 more fraud attempts while holding up 472 fewer good customers at identical cost, and is still not promoted, because its calibration and population stability fail checks written before the result was known.
PROFESSIONAL SNAPSHOT
Data foundations. AI systems. Architecture that ships.
Azure- and AWS-certified, with a data analyst’s rigor and a data scientist’s evaluation mindset—progressing into AI engineering and architecture.
Professional profile