I build scalable backend infrastructure and robust ML production systems designed for reliability and uncertainty handling.

• Architecting a modular GraphBuilder framework to decouple graph topology from dataset structures, enabling Graph Neural Networks (GNNs) to generalize across irregular spatial coordinate grids.
• Core Systems Optimization: Implemented kNN, Delaunay triangulation, and proximity graph generation pipelines to minimize inference variance. Reduced graph construction complexity from $O(N^2)$ to $O(E)$ for scalable training.
• CI Pipeline Architecture: Resolved runtime testing constraints by introducing lazy dataset initialization during test collection, enabling stable offline environments.
• Numerical Precision & Configuration: Investigating floating-point precision anomalies in time-handling logic and designing path-based graph resolution configurations.

• Core Architecture Documentation: Upgraded internal technical documentation for the enterprise ReferenceIndex system to improve developer velocity.
• UI System Reliability: Fixed layout constraints and component overflow issues handling highly dynamic, deeply nested label lengths.

• Auth Infrastructure Refactor: Overhauled authentication exception flows to properly classify and map internal system errors into RFC-compliant HTTP status streams (401/403/500).

The System: An AI-assisted routing pipeline built to forecast cross-border transaction optimization and routing.

• Engineering Insight: Analyzed how minute model prediction variations cascade into financially damaging errors under market spikes. Designed granular fallback validation checkpoints to absorb upstream latency and data variance securely.
• Stack: Python, Node.js, Scikit-learn, Flask, MongoDB, SQL

The System: End-to-end medical image classification platform offering deterministic streaming inference.

• Engineering Insight: Isolated prediction instabilities caused by subtle input image resolution changes across varied hardware systems. Engineered a robust, uniform tensor normalization layer to eliminate output prediction drift.
• Stack: TensorFlow, OpenCV, Streamlit, NumPy

The System: Context-grounded LLM query interface designed around deep philosophical text embeddings.

• Engineering Insight: Defeated classic generative hallucinations by implementing a strict multi-step vector retrieval strategy alongside absolute system boundaries and tailored conversation token-saving states.
• Stack: Python, LangChain, OpenAI API, Vercel

Category	Tools & Technologies
Languages
Backend Frameworks
ML & Data Infrastructure
DevOps & Cloud

I focus on structural performance and data integrity—building code that survives real-world workloads, not just ideal conditions. If you're looking for an engineer who handles systems, failure modes, and algorithmic accuracy with equal precision, let's talk.

• 📧 Email: nitianhelpdex100@gmail.com