I build production-grade AI systems focused on inference optimization, distributed ML infrastructure, agent orchestration, and scalable backend platforms.
Currently pursuing B.Tech in Electronics & Communication Engineering at BIT Mesra (CGPA: 9.0/10.0) while building systems that improve latency, throughput, reliability and deployment efficiency for modern AI applications.
- LLM Inference Optimization
- Triton & CUDA-based Systems
- AI Infrastructure Engineering
- Agent Orchestration Frameworks
- Distributed Systems
- Production ML Platforms
- Reliability & Observability
🏅 Amazon ML Summer School Scholar (Top 0.2% Nationwide)
🏅 CDAC Merit Scholar
🏅 Open Source Contributor (GSSOC)
🏅 ML Engineer @ NxtWave
🏅 ML Engineer @ Elevate Labs
🏅 AI Systems Intern @ OutriX
🏅 Cybersecurity Intern @ CDAC India
🏅 Algorithmic Trading Intern @ Lunor AI
- Engineered ML pipelines powering AI-driven educational products and adaptive learning experiences, processing learner interaction data across a platform with 131M+ coding submissions.
- Improved model development efficiency by 40% through automated experimentation, feature engineering, hyperparameter optimization, and reproducible evaluation workflows.
- Designed PyTorch training and inference pipelines for NLP and computer vision tasks; improved experiment reproducibility through structured preprocessing and automated evaluation tooling.
- Optimized inference workflows via latency profiling and batching strategies, reducing average inference time by ∼18% across 3 deployed model variants.
- Built ML evaluation harness for model validation, benchmarking and regression testing across 5 model iterations
- Built an LLM evaluation pipeline processing 1M+ records — automated scoring, regression testing and failure triage — cutting experimentation turnaround time by 30%.
- Owned ETL/ELT data workflows feeding inference benchmarking dashboards; instrumented with OpenTelemetry for end-to-end latency observability.
- Profiled high-throughput AI inference workflows,identifying 3 bottleneck stages optimized to reduce p95 latency by ∼18%.
- Built a 3-stage anomaly detection pipeline on structured network-intrusion datasets (∼50K samples): feature extraction → threshold calibration → alert triage, reducing manual review queue by ∼35%.
- Implemented distributed validation and monitoring workflows for automated anomaly scoring across multi-source security data streams.
- Developed deterministic multi-asset trading strategies using SQL-backed financial time-series datasets.
- Built backtesting systems evaluating Sharpe ratio, volatility and maximum drawdown for strategy validation.
- Implemented volatility-adjusted optimization techniques improving risk-adjusted returns and portfolio stability.
Python • C++ • TypeScript • SQL
PyTorch • Transformers • LLMs • RAG • CNNs • Agent Systems
CUDA • Triton • TensorRT • FlashAttention • Quantization • KV Cache Optimization
FastAPI • Redis • PostgreSQL • Docker • Kubernetes • AWS
OpenTelemetry • MLflow • Monitoring • Performance Profiling
AsyncIO • Event-Driven Architecture • Scheduling • Caching • Message Queues
I enjoy solving engineering problems involving:
- GPU Utilization Optimization
- Inference Throughput Scaling
- Low-Latency Architectures
- Distributed Scheduling
- Agent Systems
- AI Reliability Engineering
- Production AI Deployment
💻 GitHub
🧠 LeetCode