Production AI engineering in clinical documentation; research in trustworthy language models, retrieval, and dynamic representations. This record brings together my experience, technical contributions, research, service, and source links.
01 / SELECTED ENGINEERING
The problem, the build, the result.
The problem, my contribution, the system, and the result. A closer look at how I take AI beyond an isolated model.
LONG HEALTH · JUN 2025–PRESENT
EvalPath
From fragmented medical records to structured clinical documentation.
PRODUCTION / AI ENGINEER
The problem
Specialized medical evaluations bring together scanned records, physician–patient conversations, exam findings, and final reports. Each stage needs to connect to the next.
What I own
End-to-end platform engineering: document ingestion and indexing, asynchronous OCR, retrieval, transcription, structured report generation, and role-aware physician portals.
How I built it
Document infrastructure. AWS Lambda, ECS, S3, RabbitMQ, and Textract for ingestion and processing.
Clinical AI workflows. ChromaDB retrieval, OpenAI and Anthropic integrations, ICD-10 inference, and Whisper transcription.
Product delivery. Angular/NestJS interfaces, case assignment, four-role access control, exports, monitoring, and pipeline alerts.
Records & audio→OCR & transcription→Retrieval & extraction→Structured reports
35%higher document throughput*
50%less documentation & triage time*
*Outcomes reported in my career materials. Public baselines and measurement periods are unavailable. Employer systems and clinical data are private.
INDEPENDENT PROJECT · CREATOR
CausEval.
Do your AI tests catch missing instructions?
OPEN SOURCE / LLM EVALUATION
The problem
A passing evaluation suite can still miss important prompt instructions. Test pass rate alone does not show which behaviors the suite actually checks.
My contribution
I created CausEval to extract behavioral rules, map them to evaluations, remove one instruction at a time, and compare repeated runs. The result is an inspectable account of detected and missed changes.
Clinical AI, public safety, industry analytics, and academic research. Hands-on responsibility from experimentation through software delivery.
Jun 2025–Present01
Long Health
AI Engineer
AI platforms & full-stack systems
Built and maintained EvalPath across cloud infrastructure, document AI, retrieval, model integrations, and physician-facing workflows. Partnered with clinical and compliance teams on privacy-sensitive data handling.
Supported CSC 522: Automated Learning and Data Analysis with Prof. Thomas Price, and CSC 591/791: Real-Time AI and Machine Learning Systems with Prof. Xipeng Shen. Evaluated projects, developed grading rubrics, held office hours, and helped students with validation, model selection, quantization, and inference trade-offs.
scikit-learn · TensorRT · ONNX Runtime · ML evaluation
Jan–Jun 202304
DRDO · Centre for AI & Robotics
Machine Learning Researcher
Streaming NLP & temporal graph learning
Implemented temporal graph models in PyTorch Geometric and incremental BERT workflows for evolving text. Conducted model-design and experimental work on dynamic representations and semantic change.
PyTorch · PyG · BERT · Temporal GNNs
May–Jul 202205
Merkle
Data Science Intern
Retail ML & data engineering
Led a four-person internship team working with 10M+ retail records. Built XGBoost, LightGBM, and LSTM models; engineered ETL across 16M+ rows using PySpark, SQL, and Snowflake; delivered pricing and segmentation dashboards. The project received the Stellar Team Award.
PySpark · SQL · Snowflake · XGBoost · Power BI · Tableau
Mar 2021–Jun 202206
Manipal Institute of Technology
Machine Learning Researcher
Medical imaging & network prediction
Compared VGG-16, MobileNet, InceptionV3, and XceptionNet for pediatric bone-age assessment. Worked on LSTM/Bi-LSTM link-quality prediction and automated data, training, and API workflows.
B.Tech. minor: Big Data Analytics. Graduate coursework includes NLP, Generative AI, Machine Learning with Graphs, Privacy in AI, and Database Management Systems.
Professional certificates +
Data Science Specialization / NC State UniversityCertificate ↗
Every work is labeled by status. Publisher and author-hosted records are linked below.
PublishedIEEE SoutheastCon / 2026
TrustBench: Benchmarking Trustworthy Large Language Models with Cost and Stability Metrics
Deployment-oriented evaluation of LLM behavior, repeatability, confidence, and cost.
Compares five proprietary models across ambiguity, safety, and multi-step tasks with repeated trials. Distinguishes stable behavior from verified task correctness and examines how pricing changes value comparisons.
Learning Dynamic Representations in Large Language Models for Evolving Data Streams
Incremental language modeling and dynamic contextualized word embeddings for streaming text.
Coauthored research on incremental BERT and dynamic graphs for changes in language over time. My primary contribution was model design and experiments. Springer lists the author as J. Aravinda Raman.
SDN-Based Multipath Data Offloading Scheme Using Link Quality Prediction for LTE and WiFi Networks
Sequence models for link-quality prediction and data offloading in heterogeneous networks.
Studies LSTM and bidirectional LSTM predictions using signal strength and packet data rate to inform LTE/WiFi offloading. My primary contribution was model design and experiments.
Diabetes Prognosis using Machine Learning: A Comparative Analysis of Classification Algorithms
Comparison of K-nearest neighbors, random forests, and neural networks for diabetes prediction.
A comparative modeling study examining preprocessing, feature engineering, and classifier performance. Reported research performance should not be interpreted as clinical validation.
Privacy Awareness in Large Language Models: Input Regurgitation and Prompt-Induced Sanitization for HIPAA and GDPR Compliance
Input regurgitation and prompt-based sanitization in privacy-sensitive LLM applications.
Studies sensitive-information leakage and prompting strategies on synthetic inputs. These experiments examine privacy behavior; they do not establish regulatory compliance for a deployed system.
Using Transformer-Based Models to Optimize Inventory Replenishment Decisions in Dynamic E-Commerce Markets
Attention-based modeling for inventory decisions under changing demand and seasonality.
Investigates multi-head attention for historical sales, promotions, pricing, and external signals. This work is under review; no publication is claimed.
Under review No public link
ManuscriptAuthor-hosted
Improving Fairness in Visual Recognition through Feature Distillation and Adversarial Debiasing
Feature distillation and representation learning for fairness in visual recognition.
Explores an MMD-based distillation objective to reduce disparity while maintaining task performance. A manuscript, with no peer-reviewed publication status claimed.
Publication links are provided where available. Accepted work, preprints, and manuscripts are listed separately from published papers. Research summaries are concise descriptions, not verbatim abstracts.
Built a full-stack health-insurance document assistant: upload plans, retrieve relevant passages, ask questions, compare coverage, and export personalized PDF reports.
Graduate teaching assistant for Automated Learning and Data Analysis and Real-Time AI and Machine Learning Systems.
Supported project evaluation, grading rubrics, model selection, and efficient inference.
INDUSTRY RECOGNITIONStellar Team AwardMerkle internship cohort
Team recognition for the internship project.
07 / COLLEAGUE PERSPECTIVES
From the people I've worked with.
“His attention to detail, analytical insight, and dedication made him an invaluable asset to our team.”
Jason KlinkPublic Safety Professional / Co-Founder, SmartProtectDirect manager / October 30, 2024LinkedIn profile ↗Read full recommendation +
I've had the pleasure of working with Aravinda Jatavallabha as part of the SmartProtect Public Safety Solutions team, where he served as a Data Science Intern over the past six months. Aravinda's contributions to building out our predictive analytics framework for public safety have been both impactful and highly innovative. His work, especially in structuring and analyzing complex datasets, significantly advanced our ability to deliver tailored, data-driven solutions to our clients.
Aravinda brings deep technical expertise to his work, combined with a genuine commitment to understanding and solving client challenges in public safety. His attention to detail, analytical insight, and dedication made him an invaluable asset to our team, and his contributions will continue to benefit both our company and the customers we serve.
“He was a valued team member with great communication skills and someone I trusted to bring both integrity and innovation to the table.”
Ricardo BlancoCRO / Co-Founder, SmartProtectDirect manager / June 27, 2025LinkedIn profile ↗Read full recommendation +
I had the pleasure of working with Aravinda during his time at SmartProtect, where he made a lasting impact on our AI/ML initiatives. From day one, he brought thoughtfulness and a sharp technical mindset to every challenge. Whether he was building early prototypes or helping us think through how machine learning could meaningfully support public safety operations, Aravinda consistently delivered high-quality work and insightful ideas.
What stood out most was his willingness to dive deep, learn fast, and look for ways to make the work better, not just from a technical standpoint, but in a way that aligned with the mission of serving first responders. He was a valued team member with great communication skills and someone I trusted to bring both integrity and innovation to the table.
Any team would be lucky to have Aravinda on board, and I'm excited to see where his journey takes him next.
Recommendations are reproduced with light punctuation cleanup for readability. Links lead to the recommenders' profiles.
Record & source notes
Prepared from my supplied CV, existing portfolio, and linked research and project records. Accepted work is separate from published papers; preprints and author-hosted manuscripts are not presented as peer-reviewed publications. Invited reviewer roles do not establish a completed-review count. Merkle’s award is team recognition within the internship cohort.
Employer metrics are reported outcomes with no public baseline or measurement period. Source links are provided for inspection; availability and independent verification vary. The resume PDF is preserved as supplied. Publication-name variants include Aravinda Raman Jatavallabha, Aravinda Jatavallabha, Aravinda Raman, and J. Aravinda Raman.
