ML Engineer

BHUVAN KARTHIK CHANNAGIRI

+1-857-***-**** Boston, MA **********.*@************.*** LinkedIn Personal Portfolio GitHub SUMMARY

ML engineer with a Master's in Machine Intelligence from Northeastern and research experience at Harvard, working across CV, agentic LLM systems, and end-to-end ML pipelines. Built SegFormer-B2 segmentation models with SLURM-orchestrated distributed training reaching 89.55% Pixel Accuracy, deployed Vertex AI pipelines on GCP, and architected multi-agent CrewAI systems with tool-augmented retrieval. Strong in PyTorch, JAX, MLOps tooling, and adversarial robustness for safety-critical ML. SKILLS

ML & Agentic AI: PyTorch, TensorFlow, JAX, LLMs, Computer Vision, NLP, CrewAI, n8n, LangChain, RAG, ChromaDB, Flowise Data Engineering & MLOps: GCP, AWS, Vertex AI, Docker, Kubernetes, PySpark, Apache Airflow, MLflow, DVC, Fast API, Git Programming & Web Integration: Python, SQL, MongoDB, C++, C, React.js, Flask, HTML/CSS, MATLAB, Shell Scripting, REST APIs Data Visualization & Analytics: Tableau, Power BI, Grafana, Looker, Seaborn, Matplotlib, Plotly, Streamlit, Pandas, NumPy Bioinformatics & Image Analysis: CellPose, FIJI (ImageJ), Napari, Imaris, µSAM, Stata, DeepCell, Nextflow, BioPython, CyCIF WORK EXPERIENCE

Machine Learning Research Assistant, The Rowland Institute at Harvard, Cambridge, MA Jan 25 – Jul 25

• Engineered high-throughput image-analysis pipelines using ML and industry-standard tools (CellPose, FIJI/ImageJ, Imaris, µSAM for Napari), incorporating customized thresholding strategies to elevate segmentation accuracy to 90%

• Architected GCN models on large-scale graph datasets (50K+ nodes) to capture complex spatial relationships expertise directly transferable to healthcare for modeling patient biomarker and cellular interaction networks, and proteomics data

• Designed graph-based GAN frameworks to optimize node-importance weighting and evaluate performance impacts Graduate Assistant (Machine Learning), Northeastern University, Boston, MA Verifiable Machine Learning (EECE/CS 7268) Aug 25 – Dec 25

• Collaborated with the course professor to develop and validate adversarial robustness experiments on medical imaging datasets, applying formal verification techniques to safety-critical ML deployment scenarios in healthcare and aviation

• Engineered adversarial-attack modules (UAP, PGD, FGSM, Adversarial Patch) across 25,000 LC25000 histopathology images, and implemented certified defenses using JAX and AutoLiRPA via bound propagation and convex relaxations

• Established Vision Transformers as adversarial defenses, exploiting global self-attention to sustain 92% of baseline accuracy under PGD, UAP, and adversarial-patch attacks, limiting degradation to 8% against deterministic safety benchmarks LLM-Based Dialog Agents (EECE/CS 7398) Aug 24 – Dec 24

• Architected CivicCrew, a 5-agent CrewAI pipeline, that converts vague civic requests into agency-routed action plans, pre- filled forms, and auto-drafted appeal letters with tool-augmented web retrieval and structured JSON outputs

• Engineered a PostgreSQL state and observability layer tracking per-agent telemetry, source provenance, and freshness scores; implemented a critic-revise loop to reduce hallucinated steps and surface confidence-scored recommendations

• Built an LLM-as-judge evaluation harness over hand-curated civic tasks measuring procedure accuracy and citation grounding, iterating agent prompts and tool selection to improve end-to-end success rate and reliability PROJECTS

Learning from Synthetic Humans: Small-Data Human Part Segmentation, [GitHub]

• Engineered a part-segmentation pipeline using SegFormer-B2 with SLURM-orchestrated distributed training on a constrained SURREAL subset

• Achieved 89.55% Pixel Accuracy and 50.20% mIoU in 9 epochs on 700K samples from the 6.5M-image SURREAL dataset

• Optimized fine-tuning with transfer learning and Class-Balanced Cross-Entropy loss to sharpen boundary delineation Ozone Level Detection (End to End Machine Learning Pipeline), [GitHub]

• Developed and deployed a machine learning pipeline for ozone level detection on Google Cloud Platform, integrating services such as GCS, GCP, Airflow, and Vertex AI, resulting in a 30% improvement in deployment efficiency

• Automated data processing and model training with Airflow DAGs, incorporating error handling and data quality checks, which decreased data processing time by 25% and improved model accuracy by 15%

• Containerized the model using Docker and optimized it through hyperparameter tuning and cross-validation, improving prediction reliability by 20% and reducing false positives EDUCATION

Northeastern University, Boston, MA Dec 2025

Master of Science in Electrical and Computer Engineering, Machine Intelligence R.M.K. Engineering College, Chennai, India May 2023 Bachelor of Engineering in Electronics and Communication Engineering

Contact this candidate