AnushaVyas
I teach machines to see. Deepfake detection, semantic search, real-time gesture recognition. Models that ship, not just notebooks that run.
01AI & ML EngineeringProduction RAG and LLM systems serving 8,000+ users across 4M internal documents.
02Data Science & AnalyticsBuilt Spark ETL and forecasting pipelines for high-volume telemetry and predictive maintenance.
034× AWS CertifiedCredentials spanning cloud foundations, microservices and CI/CD, data pipelines, and security.
04Python · SQLCore tools for data engineering, feature preparation, analytics, and model development.
05LLM & Model EvaluationBenchmarked six open-source LLMs on latency, context fidelity, and task quality.
06MS Computer ScienceThe University of Texas at Arlington · May 2026.
01
Experience
AI/ML Engineer
General Motors · Hendersonville, TN
- Architected an enterprise Retrieval-Augmented Generation system using LangChain, FAISS, and GPT-4o with semantic vector search across 4M internal documents, cutting engineer query resolution time from 4 hours to 12 minutes for 8,000+ users.
- Fine-tuned domain-specific LLMs on 2.3M automotive maintenance records using LoRA adapters and PEFT on AWS SageMaker, scaling adoption across 6 product lines and reducing manual diagnostics effort by 1,400 hours per month.
- Deployed production multi-agent orchestration with LangGraph and CrewAI using function calling and tool use to automate vehicle diagnostics reporting, compressing report generation from 48 hours to 3 hours across North American operations.
- Established end-to-end MLOps infrastructure with MLflow, Docker, and Kubernetes, delivering CI/CD pipelines with canary deployments, A/B testing, and model monitoring that shortened release cycles from 3 weeks to 4 days.
ML Engineer
Persistent Systems · India
- Developed a computer vision defect detection pipeline with PyTorch and YOLOv8 for manufacturing quality control, scaling inspection throughput to 3,200 units daily across 4 global factories.
- Constructed production NLP models (BERT, RoBERTa, Hugging Face Transformers) for named entity recognition and contract classification, processing 18,000 legal documents monthly and compressing analyst review from 6 days to 9 hours.
- Automated model retraining pipelines using Apache Airflow and MLflow on Azure Machine Learning with drift detection and monitoring alerts, cutting time from drift detection to redeployment from 11 days to 18 hours.
- Benchmarked 6 open-source LLMs (LLaMA 2, Mistral, Phi-2) across latency, context fidelity, and evaluation criteria, producing a technical scorecard that became the team's standard LLM qualification process.
- Built a time series forecasting pipeline using LSTM networks and Temporal Fusion Transformer on Azure ML to predict equipment failure 72 hours in advance, eliminating 420 hours of unplanned downtime annually.
Data Engineer
Dell Technologies · India
- Designed a fault-tolerant Apache Spark ETL pipeline ingesting 4.8TB of daily device telemetry from 200,000 enterprise storage units into Snowflake, enabling downstream predictive maintenance models across 30 countries.
- Built automated data quality validation with Great Expectations and dbt, intercepting 98 schema violations monthly before propagation to the ML feature store and protecting 12 downstream production models.
- Implemented a feature engineering framework with PySpark and Apache Hive serving 40 pre-computed ML features at 25ms latency for real-time anomaly detection.
- Orchestrated Apache Kafka streaming pipelines processing real-time telemetry at 120,000 events per second, enabling near-real-time anomaly alerting with sub-5-second end-to-end latency.
- Optimized Spark job performance through partition pruning, broadcast join tuning, and dynamic resource allocation on YARN, reducing average batch runtime from 3.4 hours to 47 minutes.
02
Projects
Autonomous Multi-Agent
Research Assistant
Production multi-agent agentic system using LangGraph and LlamaIndex with Tavily web search and RAG-powered synthesis across 50,000 indexed documents, enabling autonomous multi-step tool use and function calling. Sub-8-second query latency via parallel workflows, Pinecone vector caching, and streaming inference over FastAPI with SSE, sustaining 99.4% uptime under 200 concurrent users.
Real-Time Financial News
Sentiment Pipeline
End-to-end streaming NLP pipeline using Apache Kafka and FinBERT classifying financial news sentiment at 2,000 articles per minute across 14 live sources, surfaced on a Streamlit dashboard with sub-2-second publication-to-score latency. A batch inference layer with PyTorch and Redis caching labelled 840,000 historical records, lifting downstream F1 from 0.74 to 0.89.
DeepFake Detection
Vision-KAN + Vision-LSTM
Dual-encoder hybrid pairing a 24-layer Vision-KAN transformer for spatial artifact detection with a Vision-LSTM modeling sequential dependencies across patches. Trained on 100k+ DFDC images: 96.5% accuracy, 0.981 ROC-AUC, 0.94 F1, 3.9% EER.
Cloud Resume Analyzer
Code ↗Semantic matching over 3,000+ resume PDFs using S-BERT embeddings (all-MiniLM-L6-v2, 384-dim) via Hugging Face. Ranks candidate-to-role fit by cosine similarity, served from a Flask backend on AWS with S3 storage.
Hand Gesture Recognition
Code ↗Real-time vision system at 30 FPS with MediaPipe and OpenCV, classifying 10+ gestures from 21-point landmark geometry at 92% accuracy. Temporal smoothing and dynamic thresholding cut false positives under continuous use.
03
Certifications
04
Education
UTA
The University of Texas at Arlington
Master of Science (MS), Computer Science
Aug 2024 – May 2026
◇ Web Development, Engineering and +7 skills
GTU
Gujarat Technological University (GTU)
Bachelor of Engineering (BE), Computer Engineering
Aug 2020 – May 2024
◇ Web Development, Engineering and +7 skills
05
Coursework
Data Analysis & Modeling Techniques
Design & Analysis of Algorithms
Artificial Intelligence
Software Engineering
Data Mining
Machine Learning
Cloud Computing
Web Data Management
Computer Vision
Distributed Systems
Data Structures
Database Management Systems
Operating System
Probability and Statistics
Big Data Analytics
06
Campus engagement
Graduate Assistant
ISEC – International Student & Scholar Services, UTA
Aug 2025 – May 2026
Managed PEAK Peer Empowerment (7 student leaders) and ISPB (8 members) for intercultural programming. Supported international students with admin processes, event coordination, and cross-functional collaboration.
Operations & Logistics Lead
Datathon Team Lead · UTA
2025
Led a 5-member datathon team, managing project timelines, task allocation, and team coordination to deliver competition results under strict deadlines.
External Relations Representative
Student Government Association – Executive Branch, UTA
2025
Represented student interests through external relations and executive committee initiatives. Collaborated with student leaders on campus advocacy and leadership development.
Student Assistant
Admissions Office · UTA
Mar 2025 – Aug 2025
Assisted prospective students with admissions inquiries, supported office operations, maintained records, and provided front-desk administrative support.
Council Member
Engineering Student Council · UTA
2025
Organized the NAESC Conference 2025 at UTA. Coordinated campus visits to American Airlines and Herzog for industry exposure and professional development.