Browser-based benchmark comparing CPU (TypeScript) vs GPU (WebGPU) performance on a data pipeline: filter → normalize → transform → aggregate. Tests identical workloads across input sizes (10K–10M …

A verified expression compiler with WASM + CUDA backends and a Lean 4 correctness proof. Zero sorry. Zero axioms. Kernel-checked.

How context quality affects LLM fraud detection — Mistral AI eval experiment with Pass³, DeepEval, LLM-as-judge, and counterfactual sensitivity

An experiment proving that LLM output quality is determined by context quality that covers retrieval degradation, MVCC snapshot isolation, feedback loops, and token budget prioritization.

Hi there, I'm Sweta 👋

Governance-Aware LLM Gateway: FastAPI + Claude with context scoping, risk flags, confidence thresholds, and audit logging

Real-time log ingestion and anomaly detection pipeline using Kafka, Apache Flink, Vector, and MinIO

Build AI Agents using Snowflake Semantic Views

My curated solutions to most asked LeetCode problems

A modern analytics data stack that uses dbt models and cube.js schema to create semantic layers to sync automatically with Tableau, Power BI, Superset.

Build an app with Flink Sketch maven modules to extract data structures like CountMinSketch, hyperLogLog, bloom filters, top K sketches

Use Kafka Streams serializers and deserializers to perform stateful operations with error handling

A collection of projects from predictive analytics, classifiers, tuning models, model comparisons, performance evaluations over public datasets.

Working with Iceberg tables using dlt libraries