About • Current Focus • Featured Projects • Engineering Stack • GitHub Overview • Contact
Hey, I'm Aadrit
class Engineer:
def __init__(self):
self.name = "Aadrit"
self.primary_language = "Python"
self.focus_areas = [
"Information Retrieval & RAG",
"Systems Architecture & Cache Mechanics",
"Reinforcement Learning & Policy Dynamics",
]
self.current_builds = ["SuperRAG", "HemlockV2", "primordial-void"]- Context Retrieval: Building SuperRAG for multi-source document ingestion and fast query answering.
- Systems & Memory: Measuring cache hit rates and memory latency in HemlockV2.
- Reinforcement Learning: Tracking policy divergence and entropy schedules in primordial-void.
- Spatial Data: Modeling atmospheric datasets with graph networks in Weather-Model-Graphs.
| Project | Architecture & Scope | Core Technologies | Repository |
|---|---|---|---|
| SuperRAG | Multi-source context retrieval pipeline that parses, indexes, and queries unstructured documents. | Python RAG Search |
Code |
| HemlockV2 | Systems testbed for analyzing CPU cache mechanics, throughput, and memory alignment. | Systems Memory Benchmarking |
Code |
| primordial-void | Reinforcement learning environment analyzing policy stability and gradient dynamics. | Python PyTorch RL |
Code |
| Weather-Model-Graphs | Graph representations for spatio-temporal atmospheric modeling and topological inspection. | Python Graphs Data Viz |
Code |
| Domain | Tools & Technologies |
|---|---|
| Languages & Scripting | Python, Bash |
| Machine Learning & Search | PyTorch, Information Retrieval, Vector Embeddings |
| Systems & Infrastructure | Linux, Docker, Redis, Git, Cache Hierarchy |
| Quality & Automation | GitHub Actions, Pre-commit, Static Analysis |
Architecture & Implementation Notes
- Retrieval Architecture: Evaluating vector indexing trade-offs against sparse lexical retrieval for context-augmented workflows.
- Memory Layouts: Investigating contiguous buffer allocations versus node-pointer graphs to minimize CPU cache thrashing.
- RL Stability: Focusing on deterministic seed reproduction, variance reduction techniques, and clipping thresholds.
