A RocksDB/LevelDB-inspired key-value storage engine built from scratch in Rust with zero runtime dependencies. It runs on an LSM-tree architecture with a WAL, immutable sorted SSTables with sparse indexing, and size-tiered compaction. Crash recovery through manifest checkpoints and WAL replay enforces fsync ordering and atomic file installation, verified by property testing against a reference model.
A multi-tenant SaaS that turns raw contracts into structured, reviewable data. An LLM extraction pipeline pulls fields from PDF and DOCX with per-field confidence scoring reaches 97.9% field-level accuracy on a validated 30-document test set. Tenants are isolated at the ORM layer with tenant-scoped queries, verified for zero cross-tenant leakage through automated integration tests, and secured with five independent HMAC token systems.
An end-to-end pipeline that scans React apps, generates WCAG fixes, verifies them independently, and opens GitHub pull requests. Orchestration runs on a LangGraph deterministic state machine with a crash-recoverable async pipeline backed by PostgreSQL state and Redis/RQ workers. Fixes are validated across 2 independent React/Vite codebases with multimodal visual-regression comparison, and a confidence-gated localization check defers to human review instead of guessing.
RCSB Protein Data BankScientific Software Research AssistantJul 2026 — Present
New Brunswick, NJ
Orchestrated a new MCP server tool exposing the PDB search API to LLM agents, enabling multi-service AND/OR queries in a single call by composing the filter nodes from the 7 extended tools through a custom validation layer.
Rectified 2 silent data corruption bugs in the MCP server's result-grouping logic by tracing parameter flow against the tool's own docstrings, preventing the formation of invalid API parameters and user configurations.
Rutgers AI & Data Science CollaboratorySummer Research Fellow · CSI SURFMay 2026 — Jul 2026
New Brunswick, NJ
Redesigned a dictionary-driven column-typing classifier in py-mmcif, an open-source Python package, adding automatic data type detection to cut molecular structure file sizes by 27%, submitted upstream as a pull request.
Optimized floating-point binary encoding in py-mmcif's compression pipeline, cutting compressed file sizes by up to 76% across 37 varied PDB structures, working with a team of 4 researchers on the encoding chain design.
27%
File size reduction
76%
Compression improvement
37
PDB structures
04The person behind the work
Off the Clock
I'm Hriday, a Computer Science and Economics student at Rutgers University–New Brunswick. I enjoy building software where systems engineering, AI, and practical problem solving intersect.
Outside of development, I'm usually gaming on my PS5, reading about space, or disappearing into another technical rabbit hole.
Space → controlsThe same instinct that pulls me toward orbital mechanics pulls me toward a well-designed system: a small set of inputs, precise feedback, and a state you can always read.
