Shel Burkes, PhD
Scientist working on whether AI measures everyone equally
Summary

Scientist with a PhD and 11 years of data science and computational research, working on fairness in measurement: whether a skin, color or language measure means the same thing for everyone it is applied to, and how to test that with AI evaluation. Independent work shows where common measures break. In iridis, skin-tone measurement error traces to the camera rather than the tone scale, and a per-device color correction reaches ρ 0.915. In lambent, a radiance metric's specular threshold sits 5.3 times higher on the lightest skin type than the darkest. In veridian, an LLM screen turns a one-line topic into a mapped literature corpus: on hair fiber science it kept 1,199 of 3,205 papers, all 20 sampled inclusions in scope, and found Afro-textured hair in only 2.7% of them. Before that, four years as sole data scientist at a consumer health-sensing startup, building computer vision models that ran on in-store devices, and applied AI work on a team in a pharmaceutical animal-health division, including an LLM classification pipeline and its regression tests. Mentors junior data scientists, has trained researchers with no computational background to run their own analyses, writes production software including a multi-tenant pricing API, and publishes negative results rather than shipping weaker systems.

Experience
Boehringer Ingelheim Animal Health, via Data Science Talent
2025 to 2026
Principal Applied Scientist (Contract)
    On the applied AI team in the Global Animal Health division. Python, PyTorch, TensorFlow, Databricks, Snowflake, GitHub Copilot.
  • Contributed to technical scoping and solution architecture for applied AI engagements, drafting conceptual frameworks and first implementation approaches
  • Built model prototypes and early implementations in Python, PyTorch and TensorFlow, then handed them to engineering teams for production development
  • Worked with client teams across commercial, corporate affairs and analytics to translate business needs into scoped modeling objectives
  • Worked within regulated pharmaceutical data-access processes, coordinating permissions so modeling followed internal governance standards
  • With the team, built an LLM-based misinformation detection pipeline on Databricks over social media data in Snowflake: LLM classification of posts through the company's internal gateway of vetted LLM APIs, clustering and trend detection over flagged content, and triage output for review in Sprinklr. Delivered as a proof of concept in 60 days
  • The pipeline gave corporate communications earlier signals of misinformation about high-visibility animal health products
  • Helped define success metrics and regression tests for classification with no single right answer, so changes could be checked before handoff
  • Built a financial potential model for livestock and farming customers, estimating purchasing capacity across major customer portfolios to guide commercial targeting and outreach
  • Mentored two junior data scientists on modeling techniques, code structure and maintainable practices
  • Contributed to project planning, requirements, documentation and cross-team communication
Confidential consumer health-sensing startup
2021 to 2025
Senior Data Scientist (Contract)
    Sole data scientist for the startup, owning ML research, measurement methodology and analytics infrastructure from scoping to handoff. Part-time until 2022 alongside Syngenta.
  • Built three anomaly detection models, including a UV-mode model, in TensorFlow/Keras and PyTorch, trained on SageMaker and run in production on in-store devices for a global prestige beauty retailer and a Fortune 500 pharmacy group
  • Created image-based measures of skin radiance, an attribute that had existed only as a qualitative descriptor, validated against expert grading
  • Conceived a skin-tone classification system on the startup's internal image data, later offered to clients as a paid service
  • Designed shade-matching logic for a luxury beauty house that ranks products by perceptual color distance
  • Scoped exploratory models for a national beauty retailer and a national pharmacy chain
  • Owned A/B tests and causal inference studies from protocol design to analysis, isolating the effect of product features on user behavior
  • Defined core product health metrics (engagement, retention, conversion) used in product, engineering and executive reporting
  • Segmented customers with K-means, PCA and UMAP, including geolocation analysis, to inform product decisions
  • Built Plotly Dash and Tableau dashboards and turned findings into development priorities for clients and non-technical stakeholders
Syngenta
2021 to 2022
Data Scientist
  • Built predictive models on haplotype data to identify traits and optimize product performance based on environmental and genetic factors, combining biological domain reasoning with ML modeling end-to-end
  • Designed and deployed scalable bioinformatics workflows for trait-based prediction and optimization, integrating outputs into relational databases
  • Collaborated cross-functionally across the data science lifecycle — data wrangling, exploratory analysis, hypothesis testing, prototyping, validation, and deployment
  • Launched a decision-making analytic platform as Product Owner on a small blended team, owning the product vision and backlog using Agile and Scrum methodologies
  • Communicated complex analytical work to technical and non-technical stakeholders throughout project execution
Cooper Lab, North Carolina Research Campus
2020 to 2021
Postdoctoral Research Fellow
    Multicultural Postdoc Fellow
  • Owned the lab's analytical pipelines for large-scale NGS datasets in Python across UNIX and cloud/HPC environments
  • Developed pipelines for transposable element polymorphism detection and structural variant association analysis in tomato, linking insertion presence/absence to phenotypes
  • Mentored the lab's graduate and undergraduate researchers: wrote internal documentation, ran pipeline walkthroughs, and took students with wet-lab backgrounds and no computational experience to running their own NGS analyses
  • Maintained code documentation and data stewardship so datasets stayed reproducible across ongoing research
Schlueter Lab, University of North Carolina at Charlotte
2015 to 2020
Graduate Researcher & Lab Instructor
    GAANN Fellow (Graduate Assistance in Areas of National Need).
  • Conducted dissertation research on repeat landscapes in Avena (oat), building analytical pipelines in Python and Linux over large Illumina and PacBio sequencing datasets
  • Integrated data from NCBI and GenBank into large-scale genomic processing workflows
  • Turned the repeat-discovery work into RepBox, an open-source tool published in BMC Bioinformatics (2023)
  • Taught the lab for a graduate course, Biological Basis of Bioinformatics (BINF 8100), across three Spring semesters at 10–15 students per term
  • Served as teaching assistant for Biological Basis of Bioinformatics
Education
Publication
Certifications
Selected projects
iridis.
2023– · Results committed
Tested whether image-derived skin color tracks Fitzpatrick type and Monk Skin Tone on open dermatology images, and found much of the image error is the camera, not the scale. A color matrix per device and dermoscope mode, fitted on held-out patients, recovers a colorimeter at ρ 0.915 and cuts the capture share of variance from 40% to 5.8%.
Read the write-up →
lambent.
2023– · Results committed
Skin radiance metric from multi-region image features, developed independently on public data. Re-validated on 1,816 public dermatology images and rebuilt into a capture-robust, tone-neutral score (tone dependence ρ² 0.310 to 0.002).
Read the write-up →
veridian.
2025– · Results committed
Literature review accelerator for any field: from a one-line topic and an inclusion statement it drafts searches, snowballs citations, screens every abstract with an LLM and maps the result. On human hair fiber science it screened 3,205 candidates to 1,199 in 5.7 minutes, with all 20 sampled inclusions in scope and 20 of 24 references of an independent review found.
Read the write-up →
recolo.
2026– · Results committed
Designed memory for LLM agents (episodic and semantic stores with decay, salience and consolidation) and fixed the LongMemEval protocol before scoring. Plain retrieval over the same memories won (0.73 against 0.32 as shipped), so ablations and four more pre-registered protocols traced why and tested compaction. A redesign around what users add, revise and retract then beat both plain retrieval and rolling compaction on Memora at the same token budget (0.28 against 0.18).
Read the write-up →
noul.
2026– · Results committed (find only)
Local code search that keeps code on the machine: BM25 and an embedding model shortlist candidates and a cross-encoder reranks them. On 35 labeled queries across two codebases, every single-file answer ranks in the top five at a tenth of brute-force cost; a strict every-correct-file metric showed only 4 of 9 multi-file answers complete, and the default stays unchanged until a fix holds on a held-out codebase.
Read the write-up →
menhir.
2026– · Implemented
Co-owned training platform for athletes and coaches, heading into open beta, built under heurology with a coaching partner who owns the training content and engine logic. Designed and built the application from the ground up (screens, sign-in, data layer, AI wiring), now working alongside an associate engineer. A multi-provider LLM pipeline generates programs on top of 13 deterministic methodology engines, and every program is validated before commit. Invite-based accounts, tiered billing and an offline PWA; about 129K lines of code, 92 test files, 1,954 tests.
See the work →
mara.
2026– · Shipped
Designed and built, under heurology, a multi-tenant pricing API for the moving industry, in production as a client platform's pricing service: one survey endpoint estimates weight from job-site photos, routes the move, recommends a crew and returns priced line items. Per-tenant rates and hashed, scoped API keys; 10 test modules.
All projects →
Skills
Languages
PythonSQLTypeScriptBash
ML and statistics
PyTorchTensorFlowscikit-learnComputer vision (CNNs)Recommendation and rankingClustering and dimensionality reduction (K-means, PCA, UMAP)A/B testing and causal inferenceStatistical modeling
LLM systems
LLM classification pipelinesMulti-provider LLM pipelines with output validationRetrieval and reranking (BM25, embeddings, cross-encoders)Agent memoryEvaluation design and pre-registered protocolsLLM APIs (Anthropic, OpenAI, Groq)Claude Code (skills, subagents, hooks)GitHub Copilot
Data and deployment
DatabricksSnowflakeAWS (SageMaker, Athena)APIs and SDKsPlotly Dash and TableauLinux and HPCGit
Domains
Regulated pharmaceutical dataConsumer health and beautyGenomics and crop breedingBiomedical imaging