Alvin Alias | Data Scientist

Data scientist 12 years in mechanical engineering across subsea and HVAC MS Data Science student, University of Washington

Seattle, WA

Engine hardwaregeometry / forces / heat

Sensor data + predictiondegradation / remaining life

The range

A wider range, built on engineering experience.

Explore a capability to see the work behind it.

Machine learning 6 skills

Domain-informed feature engineering

CMAPSS Turbofan RUL HVAC Building Anomaly Scoring Industrial Failure Classification
XGBoost and classification

CMAPSS Turbofan RUL Industrial Failure Classification
SHAP interpretation

CMAPSS Turbofan RUL HVAC Building Anomaly Scoring Industrial Failure Classification
Anomaly detection

HVAC Building Anomaly Scoring
NLP and fine-tuning

Maintenance Work-Order NLP
RAG and retrieval evaluation

RAG Engineering Assistant

Statistics & inference 3 skills

Time-series forecasting

Energy Demand Forecasting
Causal inference

Causal Field Analytics
Model evaluation and validation

Energy Demand Forecasting RAG Engineering Assistant Industrial Failure Classification Retail Returns Intelligence

Deployment & APIs 1 skills

API development

CMAPSS Turbofan RUL Retail Returns Intelligence HVAC Building Anomaly Scoring Industrial Failure Classification Maintenance Work-Order NLP RAG Engineering Assistant

Programming & data tools 2 skills

Python ML stack

Energy Demand Forecasting HVAC Building Anomaly Scoring Industrial Failure Classification
SQL

Supply Chain Cost Intelligence

Engineering domain: manufacturing & reliability 4 skills

Simulation and validation

Centurion Rheem
HVAC product development

Daikin Rheem HVAC Building Anomaly Scoring
Reliability analysis

Daikin Industrial Failure Classification Causal Field Analytics
Engineering cost analysis

Daikin Supply Chain Cost Intelligence

Leadership & delivery 2 skills

Cross-functional delivery

Daikin Rheem
Technical communication

Baker Hughes (via KH Consulting)Rheem Causal Field Analytics Energy Demand Forecasting

Projects

Selected data science work across industries and problem types.

Choose a project Project details

Project 01industrial

2.88M readings, 497 buildings

HVAC Building Anomaly Scoring

Ranks unusual chilled-water operation across 497 buildings using ASHRAE meter, weather, and metadata features.

91.3%IF / LOF agreement

Isolation Forest and LOF agreed on 91.3% of a 100K-row sample, a consistency check rather than failure accuracy.

Isolation ForestLOFSHAP

Honest limitCOP and delta-T are engineering proxies; without equipment-level sensors or failure labels, the scores cannot establish compressor degradation.

Case study Live demo Source API docs

Project 02retail / ops

14 years of PJM grid load

Energy Demand Forecasting

Compares SARIMA and Prophet with a same-hour-last-week baseline in an 84-window rolling backtest on real grid demand.

7.9%SARIMA WAPE

Beat the naive baseline (8.6%) and Prophet (10.4%), winning 43 of 84 windows.

SARIMAProphetstatsmodels

Honest limitPerformance is specific to this dataset, forecast horizon, and backtest design.

Report Source

Project 03industrial + data

778 pages of DOE, NASA, OSHA

RAG Engineering Assistant

Exact standards lookups read from 88 verified facts. Explanation questions use retrieval, Groq generation, and citation checks.

50/50factual exact match

The frozen factual set reached 50 of 50 exact matches; on a separate 33-question holdout, Groq routed 29 and GPT-4o-mini routed 31.

RAGroutingInspect AIvalidation

Honest limitThe exact-match result belongs to a frozen 50-query set. New phrasing can miss the facts lookup and fall through to synthesis.

Case study Evaluation Live demo Source

Project 04industrial

NASA FD001 simulated fleet

CMAPSS Turbofan RUL

Predicts remaining useful life from rolling statistics over documented CMAPSS sensor channels. SHAP shows which inputs are associated with each estimate; it does not establish a causal fouling mechanism.

15.85RMSE

Improved this project’s Ridge baseline from 17.47 to 15.85 RMSE.

XGBoostSHAPFastAPI

Honest limitThe result is specific to this repository’s FD001 protocol and is not directly comparable with published scores until preprocessing and scoring are aligned.

Case study Live demo Source API docs

Project 05industrial

synthetic work orders

Maintenance Work-Order NLP

Classifies maintenance work orders, retrieves similar cases, and extracts structured fields. The live route uses rules, while a separate GPT extraction track was evaluated offline.

94.4%LoRA F1

On one synthetic 600-record holdout, LoRA reached 94.4% macro F1 with 1.1% trainable parameters and a 2.8 MB adapter.

DistilBERTLoRArule extractionoffline GPT eval

Honest limitTraining and holdout text share synthetic generator templates, and the live extraction route does not use the offline GPT track.

Case study Live demo Source

Project 06industrial + retail

synthetic panel with known truth

Causal Field Analytics

Tests propensity score matching, difference-in-differences, synthetic control, and Cox survival on a synthetic field panel with a planted hazard reduction.

0.812Cox HR on synthetic panel

Cox estimated 0.812 against a planted hazard ratio of 0.85; PSM was also useful, while the other methods exposed diagnostics.

DoWhylifelinesstatsmodels

Honest limitThe synthetic panel demonstrates method diagnosis, not a real field effect; DiD is imprecise and synthetic control fails its pre-fit diagnostic.

Report Source

Project 07retail / ops

100K federal-award slice

Supply Chain Cost Intelligence

Uses SQL window functions to screen a capped 100K-record slice of federal contract awards for supplier-cost review candidates.

1,456review flags

The screen produced 1,456 flags; the $182.5M figure is a heuristic scenario under explicit comparability and 30% re-sourcing assumptions.

SQLwindow functions

Honest limitAward amount is not unit price, so the flags identify review candidates rather than validated overcharges or recovered savings.

Report Source

Project 08retail / ops

1M+ real transactions

Retail Returns Intelligence

A retail analytics prototype spanning customer anomaly flags, segmentation, substitute recommendations, and policy simulation.

1M+rows audited

An audit of 1M+ transactions found leakage in the classifier and backtest from features that encode future history or cancellation signs.

LightGBMIsolation ForestDuckDBFastAPI

Honest limitThe predictive results are quarantined until the system is rebuilt with purchase-time labels, point-in-time features, and a new untouched evaluation.

Case study Live demo Source API docs

Project 09industrial

synthetic UCI AI4I 2020

Industrial Failure Classification

Flags machine failure from five sensor fields and explores how assumed miss and false-alarm costs change the operating threshold.

0.841PR-AUC on held-out split

PR-AUC reached 0.841 on the 80/20 split. The exploratory 0.775 threshold and $466K modeled cost come from a sweep over those same test labels.

XGBoostSHAPFastAPI

Honest limitAI4I is synthetic, the costs are scenario assumptions, and the threshold still needs validation without access to final test labels.

Case study Live demo Source API docs

Project 10data / research

IBM-supplied QPU projections

Quantum ML for CAR-T

Compares projected quantum kernels with a classical SVM for CAR-T classification, alongside a separate eight-construct live QPU experiment.

8live QPU constructs

Ran eight constructs in a separate live QPU experiment. On projected-kernel classification, PQK-SVM was +2.7 points on one 74-sample split while cross-validation F1 favored the classical model.

projected kernelsSVMIBM QPU

Honest limitThe live run was small, and the classification result used precomputed projections rather than outputs from that experiment.

Source

Experience

Roles and degrees.

EducationSeattle, Washington · 2025 to 2027

Relevant coursework completed or in progress through December 2026: statistical machine learning, applied statistics and experimental design, data management, scalable data systems and algorithms, software design, and data visualization.
Expected March 2027.

ExperienceFort Smith, Arkansas · 2021 to 2024

Led the 2023 Department of Energy efficiency transition and managed major upgrades for 2025 refrigerant regulations.
Led HVAC design changes through preproduction builds, coordinating with manufacturing teams, responding to operator and engineer feedback, and releasing revisions before later trials.
Created repeatable graphs from updated lab-test data to support my engineering analysis and advocated for using machine learning with historical test data in simulation tools.

ExperienceWaller, Texas · 2017 to 2021

Promoted through three engineering levels, then moved from sustaining engineering into new product engineering and owned the design and release of HVAC products.
Designed HVAC products through prototyping and testing, then revised and released them for high-volume production using feedback from assembly-line trials.
Analyzed BOM and supplier data, then implemented changes that delivered $2M+ in annual savings. Performed engineering qualification of a procurement-sourced alternate vendor that entered production and delivered another $3M in annual savings.
Contributed to team analyses of field-failure and production data that guided design changes with documented reliability improvements.

ExperienceKaty, Texas · 2014 to 2017

Translated client requirements into custom subsea tool designs and supported manufacturing and testing.
Used FEA to validate designs under the combined pressure and mechanical loads encountered in subsea oil-and-gas operations.

ExperienceHouston, Texas · 2012 to 2014

Collected field measurements and updated P&IDs, electrical schematics, fire-protection layouts, and plot plans under discipline-expert supervision.
The documentation formed part of a broader overseas plant-replication project.

EducationThrissur, India · 2008 to 2012 · University of Calicut

Graduated with First-Class Honors.

Contact

alvin.alias@gmail.com LinkedIn GitHub Resume

Graduating in March 2027 and open to full-time data science roles across industries.

alvinalias.com Alvin Alias · Data Scientist