Machine Learning Engineer | Real-Time Voice AI & Agentic Systems | LLM Engineering, MLOps & Data Infrastructure

Machine Learning Engineer at RSK IT International, building real-time voice agents running in production at drive-thrus, toll booths, and parking lots. MS in Information Systems, Northeastern University (August 2025). I bridge research and production: 4 peer-reviewed publications on explainable AI and adversarial robustness, open-source AI infrastructure on PyPI and npm, and production systems from streaming audio pipelines to agentic LLM workflows.

Resume: View Resume (PDF)

I build ML systems that survive contact with production. Right now that means real-time voice AI at RSK IT International: streaming audio pipelines with 80-120 ms chunking, VAD and directed-speech gating, ASR/NLU integration, and intent routing into POS, tolling, and gate-control systems, holding end-to-end turn times under ~500 ms across 3 live sites, deployed on-prem and in hybrid edge/cloud setups for 3 production customers and 2 POCs.

Alongside that, I design and ship open-source AI infrastructure focused on making LLM and agentic systems reliable: evals, guardrails, and observability. Earlier, I did AI research at Northeastern's Amal Lab (medical image segmentation, 18TB+ multi-omics processing, multi-GPU training) and clinical data science at Brigham and Women's Hospital (transformer NLP over 10,000+ pharma reports, market microstructure analysis).

Machine Learning Engineer, RSK IT International (September 2025 – Present)

• Engineered low-latency streaming audio pipelines for drive-thru, toll-booth, and parking-lot voice agents across 3 live sites (80-120 ms chunking, VAD/directed-speech gating, ASR, intent routing), holding end-to-end turn times under ~500 ms
• Architected and deployed serving topologies for 3 production customers and 2 POCs across on-prem edge and hybrid edge/cloud, with containerized ASR/NLU microservices and gRPC/WebRTC streaming
• Built a RAG-based post-sales QA helper over ~75 internal runbooks and a three-agent ticket-routing layer, cutting misrouted tickets and cross-team handoffs

AI/ML Research Assistant, Amal Lab, Northeastern University (June 2024 – August 2025)

• Engineered Med-SAM medical image segmentation for multi-modal datasets (MRI, CT, histopathology) with fine-tuned transformer architectures
• Processed 18TB+ TCGA multi-omics data on Spark and Dask for biomarker discovery across 33+ cancer types
• Trained a production skin-cancer classifier to 97.28% accuracy with ensembles and external validation across 3 datasets

Research Data Scientist, Brigham and Women's Hospital (August 2024 – December 2024)

• Built transformer-based NLP analytics (BERT, RoBERTa) over 10,000+ pharmaceutical reports linking media sentiment to regulatory outcomes
• Designed a HIPAA-compliant meta-analysis framework across ClinicalTrials.gov and PubMed with statistical modeling

Northeastern University, Boston, MA Master of Science in Information Systems (August 2025) Focused Coursework: Advanced Data Science & Architecture, Parallel Machine Learning & AI, LLM with Knowledge Graph Databases, Natural Language Engineering, AI Generative Modeling with focus in Finance

Veermata Jijabai Technological Institute, Mumbai, India Bachelor of Technology in Information Technology (June 2023)

Programming: Python, SQL, C++, TypeScript, Java, R, JavaScript, MATLAB, Cypher

LLM & Agentic AI: Claude API (tool calling, MCP), GPT-4o, LangChain, RAG, FAISS, ChromaDB, Sentence Transformers, CLIP, Prompt Engineering, LLM-as-a-Judge Evaluation, Agent Orchestration, Evals & Guardrails

Voice & Real-Time Systems: ASR/NLU Integration, Voice Activity Detection, gRPC/WebRTC Streaming, Real-Time Audio Chunking, Session State Machines, On-Prem/Edge/Hybrid Deployment

Machine Learning & Deep Learning: PyTorch, TensorFlow, Keras, Scikit-learn, XGBoost, LightGBM, CNN, Vision Transformers, GANs, Reinforcement Learning (PPO, DQN), Distributed Data Parallel, Multi-GPU Training, Optuna

Data Engineering: Apache Spark, PySpark, Kafka, Airflow, dbt, Dask, Stream Processing, ETL/ELT, Great Expectations, Medallion Architecture

Databases & Cloud: PostgreSQL, MySQL, MongoDB, Neo4j, Snowflake, BigQuery, Redis, Supabase, AWS (EC2, S3, Lambda), GCP (BigQuery, GCS, Pub/Sub)

MLOps & Production: Docker, Kubernetes, MLflow, CI/CD, GitHub Actions, Model Monitoring, Prometheus, Grafana, Terraform, FastAPI, gRPC, Package Publishing (PyPI, npm)

Statistics & Experimentation: A/B Testing, Bayesian Inference, Sequential Testing, Multi-Armed Bandits, Causal Inference, Time Series (ARIMA, GARCH), Monte Carlo

Quantitative Finance: Options Pricing (Black-Scholes, Heston, Monte Carlo), Portfolio Optimization, Vectorized Backtesting, Statistical Arbitrage, VaR/Expected Shortfall

• Analysis of Explainable AI Methods on Medical Image Classification - IEEE ICAECT 2023
• Adversarial Attacks and Defenses for Skin Cancer Classification - IEEE ICONAT 2023
• Intrusion Detection: A Deep Learning Approach - IEEE ICEEICT 2023
• Image Captioning Using Transformer: VisionAid - IRJET 2022

I don't just build models, I build the entire system around them: streaming ingestion, serving, evals, monitoring, and the deployment that makes it real. I've published research AND shipped production code that paying customers run every day. I understand both the math and the engineering.

• Email: joganivinay@gmail.com
• Portfolio: https://vinayjogani14.github.io/
• LinkedIn: linkedin.com/in/vinayjogani
• X: @jogani_vinay
• Google Scholar: View Profile
• Scopus ID: 58030923600
• ORCiD: 0009-0005-9568-5747