ML Engineer · Bengaluru
Final-year CS Engineering student specializing in AI/ML at Ramaiah Institute of Technology. I build retrieval systems, ML pipelines, and full-stack apps — with a focus on shipping.
2027
Graduating
AI/ML
Specialization
MSRIT
Bengaluru
About
I'm a final-year Computer Science Engineering student at MSRIT, Bengaluru, specializing in AI/ML and graduating in May 2027. I build production machine learning systems end to end — data, training, evaluation, and serving — and care about the parts most portfolios skip: monitoring, drift detection, and continuous delivery. I'm targeting ML Engineer roles where research meets real systems.
Skills
Languages
Python · C++ · SQL
ML / DL
PyTorch · Transformers · XGBoost · FAISS · LightGBM · scikit-learn
MLOps
FastAPI · Docker · MLflow · Evidently AI · Grafana · GitHub Actions
Concepts
LLM Pretraining · Scaling Laws · Ablation Studies · Sequential Recommendation · Model Drift Monitoring
Projects
Trained transformer LMs at 4 scales (1M–64M params) on TinyStories. Fit an empirical power law (R²=0.999) and compared the scaling exponent against Chinchilla's published results. Migrated runs across local, Colab, and Kaggle GPUs with custom checkpoint persistence.
Implemented BPE tokenizer, multi-head attention, RoPE, SwiGLU, RMSNorm, AdamW, cosine LR from scratch — passing Stanford CS336's full test suite. Val loss 2.01 on TinyStories. Ablations: RoPE removal caused +0.30 val loss degradation.
FAISS two-tower retrieval + LightGBM reranker reaching recall@10 0.0504 (+67% over baseline), served at p99 ~25ms and 725 req/s. Extended with SASRec (2.99M params) from scratch. Fixed train-serve distribution skew that had regressed recall@10 by 41%.
XGBoost classifier on 590K IEEE-CIS transactions (ROC-AUC 0.89). FastAPI serving with PostgreSQL logging, Evidently AI drift detection, Grafana dashboards, GitHub Actions CI/CD. Deployed to Render and Streamlit Cloud.
Contact
Open to ML Engineer roles and interesting problems in general.