Aditya Jain

Developed a Python ILP algorithm to identify reusable In-Flight Entertainment parts, reducing manual audit time by ~40%
Processed 10TB of aircraft BITE data with Databricks and PySpark to support fleet-wide predictive maintenance at scale
Designed PySpark feature pipelines with Delta Lake and Databricks Feature Store to prepare aircraft data for ML workloads
Built 4 interactive Databricks dashboards used by 15+ airlines to surface maintenance insights and repair recommendations

Implemented iterative CT reconstruction algorithms to reduce Gaussian noise, improving overall image quality by ~20%
Applied CNN models using PyTorch and OpenCV to CT scans to detect coronary artery disease and identify calcification
Accelerated CT reconstruction with GPU computing using an in-house CT simulator for faster medical image processing

Engineered Go services with MDM and REST APIs, cutting security policy deployment time by ~40% across company devices
Built Redis-backed security dashboards with GraphQL APIs tracking compliance and endpoint risk for 1,000+ customers
Implemented AI-assisted MDM workflows in Go for 15+ customers, automating security policy generation and deployment

Engineered Python AI agents with Playwright for automated app provisioning, increasing employee onboarding efficiency by 2×
Developed Vue.js and GraphQL workflows for software installation and removal across employee onboarding and offboarding
Designed self-correcting AI agents with fallback recovery, reducing manual intervention on failed provisioning by ~35%

Fixed second-order gradients for 5 MLX ops, enabling higher-order autodiff across a framework with 28K+ stars on GitHub
Resolved C++ segfaults and einsum/vmap correctness bugs across 10+ merged PRs, with all fixes validated against NumPy

Closed cross-account safety-check bypass in AWS CDK asset publishing, securing a toolchain with 5M+ weekly npm downloads
Reduced AWS CDK refactor runtime by 38% and fixed security group diff blind spots across 20+ merged CDK contributions

Fixed hybrid search ignoring pagination and query parameters in LanceDB’s Python SDK, supporting 7M+ monthly downloads
Fixed stale merge-insert delete conditions in LanceDB, preventing accidental row deletions during insert/update operations

Built a static analysis tool with Tree-sitter to map large codebases into searchable dependency graphs for AI assistants
Reduced code sent to AI assistants by ~70% using NetworkX graph retrieval to find most relevant files and dependencies
Designed an MCP server enabling AI assistants to retrieve relevant code, dependencies, and context directly from the graph

Built a GPU KV-cache manager in Rust for LLM inference with paged allocation and prompt-prefix caching across requests
Implemented LRU eviction and host-memory paging across 8 CUDA streams for efficient cache reuse under heavy concurrency
Connected Rust to CUDA C++ kernels through FFI, reaching 11K block ops/sec at 91µs average latency under cache eviction

Journey, 2023 to now