Selected engineering work
Selected systems across open-source artificial intelligence, zero-GPU edge inference pipelines, linear-complexity transformers, distributed caching architectures, and first-principles neural frameworks.
Project composition
A compact view of the primary implementation languages represented in this public project record.
Public repositories and maintained systems
Scalable community generative speech and vision pipelines on Hugging Face handling 840,000+ cumulative visits, 3,200+ combined likes, 42,000+ monthly active users, and recognized as Hugging Face Space of the Week.
Optimized CPU-only real-time facial inference pipeline achieving sub-second latency bounds; awarded National Top 10 standing and INR 1 Lakh research grant at SCNTSE 2025.
Hierarchical Linear Vision Transformer replacing quadratic attention with strict O(N) localized projections for particle track reconstruction; published in EPJ Web of Conferences and awarded Conference Gold Medal.
An open-source, OpenAI-compatible agentic web search and deep-research engine with Redis hot-caching and Qdrant semantic vector deduplication; yielded formal arXiv:2609.05463 preprint.
Closed-form mathematical algorithm transferring chromatic profiles between high-resolution imagery with near-zero latency overhead; runs fully client-side in vanilla web technologies.
Zero-dependency neural network library engineered from first principles in pure Java (JDK 8) at age 17, implementing custom matrix math and backpropagation directly in bytecode arrays.
A custom hierarchical tree data structure implemented using linked list node references in Java, optimizing pointer-based traversals, dynamic node memory management, and tree balance.