AI/ML Engineer | Deep Learning · NLP · LLM Fine-Tuning · MLOps
10+ years in engineering across automotive, healthcare & gaming · Currently pursuing MS in AI/ML @ Liverpool John Moores University
I'm an engineer with 10+ years of experience across automotive, healthcare, and gaming domains — drawing on a background in mobile and backend engineering to build pragmatic, production-ready AI/ML systems.
Most recently at Cerence AI, I spearheaded integration and delivery of GPT-based capabilities for a Land Rover program using a hybrid LLM architecture combining domain-specialized responses and RAG, resulting in ~$0.5M in revenue impact. Before that, at Outset Medical, I developed predictive models for maintenance forecasting and treatment failure detection using curated device telemetry and in-house labeled datasets. Leveraged tree-based algorithms (XGBoost) to identify anomalous patterns and enable early risk detection, improving product reliability and patient safety. This helped treatment outcomes by 30% and maintenance outcomes by 15%.
I took a deliberate break to deepen my foundations, completing a Postgraduate Programme in AI/ML at IIIT Bangalore and currently pursuing an MS in AI/ML at Liverpool John Moores University. I bring strong engineering judgment, production experience, and a pragmatic approach to applying ML in real-world systems.
🔍 Currently seeking AI/ML engineering roles where I can contribute immediately while continuing to grow through real production work.
AI / ML
MLOps & Cloud
Mobile
Liverpool John Moores University · MS Artificial Intelligence · Deadline: July 2026
| Repository | Description | Status |
|---|---|---|
| Thesis_PeFT | Comparative study of LoRA vs ReFT (PEFT methods) for mental health NLP — classification (DistilBERT on SWMH) and empathetic response generation (Llama 3.2 on Empathetic Dialogues), evaluated on F1, AUC, BERTScore, EPIC, and LLM-as-Judge | 🔄 Active |
Stack: PyTorch · HuggingFace PEFT · PyReFT · Unsloth · W&B · Presidio · Docker
| Repository | Description | Tools |
|---|---|---|
| Insurance_RAG_agent | End-to-end RAG pipeline for insurance policy Q&A — document ingestion, chunking, embedding, vector store indexing, semantic retrieval, and grounded LLM answer generation | LangChain · FAISS · OpenAI |
| Llama_Index_Query_Engine | Advanced QA agent using LlamaIndex's SubQuestionQueryEngine — decomposes complex multi-part queries into targeted sub-questions, retrieves per-sub-question context, and synthesises a unified grounded answer | LlamaIndex · OpenAI |
| Repository | Description | Tools |
|---|---|---|
| Fine-Tuning-Large-Language-Models | End-to-end fine-tuning of a pre-trained LLM on a customer experience dataset — instruction formatting, training configuration, and before/after inference comparison | HuggingFace Transformers · PEFT · PyTorch |
| LoRA_fine_tuning_on_Bloom | LoRA fine-tuning on BLOOM 1B — demonstrates low-rank adapter injection, trainable parameter reduction, and efficient adaptation of a 1B-parameter model | HuggingFace PEFT · PyTorch · bitsandbytes |
| Repository | Description | Tools |
|---|---|---|
| Healthcare_Entity_Recognition | CRF-based Named Entity Recognition on patient-doctor interaction logs — extracts disease and treatment entities using hand-crafted token features, outputs a structured disease-to-treatment dictionary | sklearn-crfsuite · spaCy · nltk |
| BERT_finetuning | Fine-tuning pre-trained BERT for sentence pair classification tasks including NLI and semantic textual similarity | HuggingFace Transformers · PyTorch |
| NMT_Encoder_Decoder_Attention_Beam | Hindi-to-English neural machine translation with GRU encoder–decoder, Bahdanau attention, and beam search decoding | TensorFlow · Keras |
| TensorFLow_RNN_language_translation | Language translation using a custom RNN encoder–decoder and loss function built from scratch with explicit GradientTape training | TensorFlow |
| semantic_processing | Notebooks covering distributional semantics, knowledge graphs, and topic modelling | spaCy · Gensim · sklearn |
Beatit-AI Churn Prediction System — AWS SageMaker (@chetnapriyadarshini-iiit)
A production-grade, end-to-end MLOps system for predicting music streaming subscriber churn, built on AWS. Spans 5 repositories covering the full ML lifecycle from raw data to monitored production inference.
| Repository | Role | Tools |
|---|---|---|
| beatit-ai-glue-redshift-tables | Bronze → Silver → Gold medallion data pipeline producing model-ready features | AWS Glue · Redshift · S3 · PySpark |
| beatit-ai-model-train | SageMaker Pipeline: preprocessing → training → evaluation → model registration with CI/CD via CodePipeline | SageMaker Pipelines · CodeBuild · XGBoost |
| beatit-ai-model-deploy | Automated staging → production endpoint deployment with CloudFormation IaC and manual approval gate | SageMaker Endpoints · CodePipeline · CloudFormation |
| beatit-ai-model-monitor | Continuous monitoring for data quality, model quality, bias, and feature attribution drift on production endpoints | SageMaker Model Monitor · Clarify · CloudWatch |
| beatit_ai_common_utilites | Shared utilities across all pipeline components | Python |
| Repository | Description | Tools |
|---|---|---|
| CodeProMLOPS | Production lead scoring system with three Airflow-orchestrated pipelines (data, training, inference), MLflow experiment tracking, and automated model promotion from Staging to Production | Airflow · MLflow · scikit-learn · Docker |
| Repository | Description | Tools |
|---|---|---|
| Telecom_Churn_Prediction | Binary classification model to predict telecom customer churn, with feature engineering and evaluation prioritising Recall and ROC-AUC to minimise missed churners | scikit-learn · pandas · matplotlib |
| PropertyPricePrediction | Regularised regression (Ridge & Lasso) to predict Australian residential property prices, identifying 12 key value drivers for a US investment firm's market entry strategy | scikit-learn · pandas · numpy |
| Bike_Rental_Prediction | Multiple linear regression model forecasting daily bike-sharing demand for post-pandemic recovery planning, with RFE feature selection and full residual diagnostics | scikit-learn · pandas · matplotlib |
| Digit-Recognizer | Handwritten digit classification on MNIST comparing a CNN against a Random Forest baseline, with Kaggle competition predictions generated for both models | TensorFlow · scikit-learn · numpy |
| Reccomendation_System | Recommendation engine implementing collaborative filtering, content-based filtering, and hybrid approaches, evaluated via RMSE and Precision@K | scikit-learn · pandas · scipy |
| Repository | Description | Tools |
|---|---|---|
| GestureRecognition | Video-based hand gesture classifier for smart TV control — compares CNN+RNN (with transfer learning) vs Conv3D architectures on 5-class gesture sequences, with a custom data generator | TensorFlow · Keras · OpenCV |
| Melanoma_Detection | Multiclass skin lesion classifier using a custom CNN trained on 2,357 ISIC dermoscopic images, with data augmentation and class rebalancing to detect melanoma | TensorFlow · Keras · Augmentor |
| Bike-Share-Neural-Network | Feedforward neural network built from scratch in NumPy — manual implementation of forward pass, backpropagation, and gradient descent — to predict bike-sharing demand (Udacity Deep Learning Nanodegree) | NumPy · pandas |
Production-quality Android applications built during the Udacity Android Developer Nanodegree (2016–2017), demonstrating REST API integration, Material Design, multi-module architecture, local persistence, and accessibility. These projects reflect the engineering foundations that underpin my current AI/ML work.
| Repository | Description | Tools |
|---|---|---|
| Capstone-Project | Open-source podcast app — ExoPlayer 2.x audio streaming and iTunes Search API for podcast discovery across categories | ExoPlayer · iTunes API · Java |
| WeatherBuddy | 14-day weather forecast app with Google Maps integration, OpenWeatherMap API, GPS location detection, and two home screen widgets | OpenWeatherMap API · Google Maps · Java |
| MovieMagic | Tablet-optimised movie discovery app with grid browsing, trailer playback, user reviews, and offline SQLite favourites via TheMovieDB API | TheMovieDB API · SQLite · Java |
| MakeAppMaterial | Material Design overhaul of a news reader — collapsing toolbars, shared element transitions, coordinated motion, and immersive imagery via CoordinatorLayout | Material Design · CoordinatorLayout · Java |
| StockHawk | Stock portfolio tracker with interactive historical charts, Yahoo Finance API, home screen widget, TalkBack accessibility support, and localisation | MPAndroidChart · Yahoo Finance API · Java |
| Build-It-Bigger | Gradle build engineering demo — multi-module Android project with free/paid build flavors, Google Cloud Endpoints backend, and AdMob ad integration | Gradle · Google Cloud Endpoints · AdMob · Java |
| Qualification | Institution | Year |
|---|---|---|
| MS, Artificial Intelligence | Liverpool John Moores University | 2025 – 2026 |
| PG Programme, AI & ML | IIIT Bangalore | 2024 – 2025 |
| Deep Learning Foundations Nanodegree | Udacity | 2017 |
| Android Developer Nanodegree | Udacity | 2016 – 2017 |
| BTech, Computer Science | Rajasthan Technical University | 2006 – 2010 |
Open to AI/ML engineering roles · Feel free to connect on LinkedIn or reach out at chetna.priya@gmail.com