Samanvya TripathiAI & ML Engineer
Education
Masters in Computer Science, Northeastern University
Graduating Dec 2024 | Boston, MA

Relevant Coursework: Machine Learning (ML), ML Operations, Natural Language Processing (NLP), Algorithms, Programming Design Paradigm, Databases

Awards
Skills
Machine Learning + Software: Python, C++, SQL, Linux, TensorFlow, PyTorch, FastAPI, Docker, Kubernetes, AWS (SageMaker, ECS, Lambda), MLflow, Regression, XGBoost, CNN, RNN, Transformers (BERT, GPT, Llama), Object Detection (YOLO), Reinforcement Learning, Hyperparameter Tuning, Transfer Learning, LightGCN, DCN, Prompt Engineering, Retrieval-Augmented Generation (RAG), DeepEval (LLM Evaluation), Phoenix Arize (LLM Observability)
Experience
Machine Learning Engineer, Bulkmagic
Jan 2025 – Present | Boston, MA
  • •Built an agentic content generation service using LangGraph with intelligent routing, complex tool orchestration, security enforcement, and end-to-end monitoring in Langfuse, enabling scalable and reliable content workflows.
  • •Engineered a FastAPI-based product summarization system with Meta Llama 3.2, achieving 99.9% API uptime through retry logic, rate limiting, and structured error handling in AWS ECS.
  • •Deployed containerized microservices on AWS ECS using Docker, reducing API latency by 30%, enabling JSON logging to CloudWatch, and integrating Phoenix Arize telemetry for LLM observability and guardrail checks.
  • •Implemented automated LLM evaluation pipelines with DeepEval, integrating hallucination detection and relevancy metrics, driving a 25% reduction in hallucinations and significantly improving AI-generated product summary accuracy.
    • Software Engineer Co-Op, Honeywell
    Jan 2024 – Aug 2024 | Fort Mill, SC
  • •Enhanced image quality in Honeywell's scanner using transformer-based models (Retinexformer, LLFormer), generating $2M annual impact through 4-second scan time reduction and 12% boost in SSIM, PSNR metrics.
  • •Applied advanced quantization techniques in TFLite and implemented optimized models on edge devices using ONNX Runtime, achieving 43% reduction in model size and 32% improvement in KPI for real-time inference.
  • •Automated ML workflows with MLflow, GitHub Actions CI/CD, and engineered data pipelines with Apache Airflow and Dask, reducing deployment time by 25% while boosting model reproducibility.
    • Machine Learning Engineer, Myelin Foundry
    Jan 2020 – Aug 2022 | Bengaluru, India
  • •Designed and optimized deep learning models for edge and cloud deployment, including Neural Architecture Search and FPGA-based solutions (Vitis AI), improving performance, adaptability, and ensuring 99.9% uptime in production.
  • •Developed advanced image/video processing pipelines with TensorFlow and OpenCV, enhancing visual quality by 10%, processing speed by 20%, and scaling to 1,000+ videos.
  • •Built and deployed REST APIs (FastAPI, Docker, PostgreSQL), cutting server load by 30% and improving response time by 50%.
  • •Established end-to-end CI/CD pipelines with Azure DevOps, Kubernetes, and Docker, reducing deployment timelines by 40% and enabling efficient parallel data processing.
  • •Deployed ML models with AWS SageMaker and Lambda, boosting scalability by 42% and automating model retraining workflows.
    • Projects
    Catalyze, MIT Hackathon

    AI-powered chemistry assistant with a multi-agent architecture for research, lab automation, and safety; integrates ChEMBL/PubChem APIs, PDF analysis, and generates automation scripts for OpenTrons (Python) and Lynx (C#).

    AI Clinical Trial Matcher

    Developed an embedding + LLM-based system for patient-trial matching; built inclusion/exclusion retrieval pipelines over 65k+ trials with ChromaDB + vector search.

    Meta-Identity, UMass Hackathon

    Developed AI-generated metaverse avatars using GPT-2, GAN, and StyleGAN, enhancing character realism and interactivity.

    NASDAQ Stock Price Prediction

    Developed predictive models for Nasdaq closing cross auction prices using CatBoost, Linear Regression, and Neural Networks, enhancing time series analysis and accuracy.

    GTA V Player Bot

    Built an AI-powered bot capable of navigating complex 3D game environments like GTA V and Witcher 3 using advanced computer vision techniques to detect, interpret, and interact with virtual scenes.