PIPELINES
Data Engineering
End-to-end ETL that turns scattered source systems into one trustworthy model, with the validation and reconciliation that regulated work demands.
PG&E · JelikaLite
data engineering × ml × public health
Ciann Amalan
Building data systems people can trust.
Data pipelines, machine learning, and LLM tools for energy, health, and sustainability measures. UC Berkeley Data Science & Public Health '27.
01.
About me
a little background, and three ways into the same problem
I'm a senior at UC Berkeley double-majoring in Data Science and Public Health, with a minor in Statistics. I specialize in building scalable data infrastructure and applying machine learning and AI to complex challenges in healthcare and energy.
Most of my work spans predictive risk modeling, advanced analytics, and data engineering, with a focus on transforming large-scale, real-world data into actionable insights that drive innovation and improve critical systems.
MODELS
Applied ML & Risk
Anomaly detection on environmental data, Monte Carlo risk simulation for the grid, and recommendation models grounded in real-world baselines.
CA Water Board · PG&E · StockVisionz
LANGUAGE MODELS
LLM Systems & Evaluation
Taxonomies, prompts, and evaluation loops that keep LLM outputs accurate, plus LLM-driven root-cause analysis that saves analysts hours.
113 Industries · CA Water Board
02.
Experience
2025 — present
Data Science Intern @ 113 Industries
Jul 2026 – Sep 2026 · Agentic AI startup
- Owned taxonomy and prompt strategy for an agentic AI consumer-insights product built on the Anthropic Claude SDK, classifying insights from Reddit, Twitter, and Instagram for enterprise users.
- Evaluated 500+ LLM outputs across 30+ test cases, refining taxonomy and prompts to cut mislabeled outputs by 25%.
- Prioritized product improvements with engineering using output-quality metrics.
Claude SDKPrompt EngineeringLLM EvaluationTaxonomy Design
Data Engineer Intern @ Pacific Gas & Electric
Jun 2026 – Aug 2026
- Engineered end-to-end ETL in Palantir Foundry integrating Employee Safety risk data from 4 source systems into a standardized enterprise risk model spanning 9 years.
- Automated safety risk monetization across Electric and Gas operations, quantifying $40M in risk exposure.
- Built schema-reconciliation pipelines against 12 forecast years of modeled risk, cutting manual reconciliation by 20%; shipped Power BI dashboards with automated refresh.
Palantir FoundryETLPower BISQL
Machine Learning Engineer Intern @ CA State Water Resources Control Board
Jan 2026 – May 2026
- Built an end-to-end ML QA/QC pipeline over 100K+ surface water chemistry records, adding StreamCat and PRISM watershed and climate features to model expected baselines.
- Isolation Forest with residual-based anomaly detection reached 89% precision and 84% recall against existing QA/QC flags.
- Integrated LangChain and Azure OpenAI for LLM root-cause analysis of flagged records, cutting investigation time by 30%.
PythonIsolation ForestLangChainAzure OpenAI
Data Engineer (Contract) @ JelikaLite
Aug 2025 – Dec 2025 · Neurostimulation research
- Developed ETL pipelines for large-scale clinical trial data in PostgreSQL, with validation, QA testing, and reconciliation for FDA regulatory compliance.
- Ran statistical modeling and power analysis in SciPy, optimizing sample size requirements by 25% and resolving discrepancies across 200+ participant records.
- Partnered with a cross-functional research team on insights supporting FDA-compliant reporting.
PostgreSQLSciPyPower AnalysisClinical Data
03.
Projects
builds, research, and coursework
ML FOR HEALTHTEST AUC 0.837
Lung Cancer Screening with ML
A lung cancer risk model on PLCO trial data, with logistic regression and SGD written from scratch in NumPy, 13 engineered features, and a parallel grid search. It reached 0.837 test AUC (vs. 0.61 with age alone), beats the NLST screening criteria at the same specificity, and was audited for AUC across sex, race, education, and smoking subgroups.
NumPyLogistic RegressionSubgroup Analysis
GitHub ↗
DEEP LEARNING94.9% ACC
Medical Imaging with Deep Learning
PyTorch models for histopathology images. A 94K-parameter CNN reached 94.9% validation accuracy on 9-class PathMNIST tissue typing, and a U-Net with skip connections, trained with Dice + BCE loss, hit 0.999 IoU on a box-segmentation task.
PyTorchCNNsSegmentation
GitHub ↗
FOUNDER · 2025 – PRESENTLIVE
StockVisionz
AI-powered stock analysis across 500+ stocks. A hybrid recommendation engine hits 82% Precision@10, transformer NLP classifies financial signals at 91% accuracy, and PostgreSQL ETL processes 1M+ records.
scikit-learnNLPPostgreSQL
Visit site ↗
CO-BUILDERVOICE AI
MeetOura
An AI portfolio-alert platform that watches your holdings and calls you when something matters. Nightly pipelines score news with FinBERT and Gemini and check RSI, MACD, and Bollinger signals, and important events trigger an outbound voice call from Oura, the in-app assistant.
Next.jsPythonFinBERTPostgreSQL
PG&E · JUL 2026STREAMLIT
Probabilistic Transmission Risk
A 3-path event tree and 100,000-iteration Monte Carlo simulation built from 17+ years of sparse outage data. It showed a ~7× jump in overload probability under projected data-center load.
Monte CarloEvent TreesStreamlit
04.
The Berkeley chapter
B.A. Data Science · B.A. Public Health · Minor in Statistics · May 2027
LEADERSHIP & TEACHING
Data 8: Foundations of Data Science
Sep 2025 – nowUndergraduate Teaching Assistant
Supports 30+ students weekly; designed learning modules from student feedback that lifted average performance by 15%.
Women in Computing & Data Science
Jan 2025 – nowVP of External Relations
Ran a 200-person networking program with 12+ tech speakers and grew membership 20% through hands-on workshops.
RELEVANT COURSEWORK
Data Structures & AlgorithmsMachine LearningData Engineering
Data Mining & AnalyticsProbabilityLinear Algebra & Diff Eq
Optimization AnalyticsRisk ModelingProduct Management
LANGUAGES & DATA
PythonSQLRJavaCPostgreSQLMySQLMongoDB
ML & AI
scikit-learnPyTorchTensorFlowPandasNumPyLangChainA/B Testing
PLATFORMS & VIZ
Palantir FoundryAWSAzureSparkMicrosoft FabricPower BITableau
05. what's next
Let's talk.
I'm looking for 2027 roles in data engineering, ML, and applied AI, especially where the work touches health, energy, or the environment.
Say hello ciann.amalan@berkeley.edu