Monil Manish Desai

Performance Evaluation of Distributed Key-Value Store Design & Eng. of Computing Systems

Architecture: Engineered a synchronous, multi-threaded Key-Value store using C++ and PostgreSQL, featuring database sharding and asynchronous replication for fault tolerance.
Performance Optimization: Implemented an in-memory LRU Cache and utilized strict Linux CPU Affinity (Pinning) to isolate workload execution on specific cores.
Bottleneck Analysis: Conducted stress testing using a custom Closed-Loop Load Generator:
- Identified CPU saturation point at ~1,960 req/s on P-cores.
- Analyzed network throughput ceilings (10.5 MB/s) and disk thrashing behavior under peak write loads.

LegalToday: AI-Driven Semantic Search Platform Software Lab

Full Stack Development: Built a responsive legal research platform using Python Flask, HTML/JS, and Tailwind CSS.
AI Integration: Integrated Google Gemini API to perform Semantic Vector Search, allowing users to find cases based on context rather than just keywords.
Key Features: Developed a real-time news feed module, a Lawyer Recommendation System based on case relevance, and automated case summarization.

Miscalibration of Deep Neural Networks (Replication) Foundations of Machine Learning

Research Replication: Reproduced the paper "On Calibration of Modern Neural Networks" by Guo et al. to analyze confidence vs. accuracy in modern architectures.
Analysis: Generated Reliability Diagrams to visualize model miscalibration and calculated Expected Calibration Error (ECE).
Mitigation: Implemented Temperature Scaling as a post-processing method to calibrate model probabilities without altering classification accuracy.

Languages	C++, Python, SQL, JavaScript, HTML5, CSS3
Systems	Distributed Systems, Linux (cgroups, CPU Affinity), Concurrency, Load Testing
Databases	PostgreSQL (Sharding & Replication), SQLite

Education