Evaluation-led
Build repeatable evaluations for data and agents. Use results to locate problems and make tradeoffs. Turn gaps into data additions, model changes, and product requirements, then retest with the same evaluation.
Working method
Use the same evaluation to decide what to change in data, models, or products
- Scientific questionDefine what users need
- EvaluationMeasure delivered results
- Root causeFind the failure stage
- ChangeData, models, or products
- RetestCheck the effect
Scientific data
Independently built and delivered a quantum chemistry library of 5 million molecules. Defined fields and acceptance criteria for molecular identifiers, 3D structures, quantum chemical properties, and computation QC. Led evaluation of the substance database, the reaction database, and external sample data, turning findings into data quality requirements.
Scientific evaluation
Used root cause analysis and ablation to locate SuperChem's chemistry reasoning bottleneck and difficulty problems in medical VQA questions. Adjusted product and question-writing direction. Responsible for scientific judgment and delivery criteria for two scientific agent task-source projects; calibrated difficulty with model runs and iterated task-writing and delivery pipelines.
Agent development
Led product definition, orchestration, frontend, backend, and evaluation for the Retrosynthesis Agent. Single-step models are open source; chemistry judgment had intern assistance. Integrating retrosynthesis into ChemMaster as its initial capability.
Scientific background
7 years combined across PhD and postdoc in computational chemistry. Long-term collaboration with experimental groups, using DFT, molecular dynamics, and microkinetic modeling to predict and explain experimental results. Published 30 journal papers, including 11 as first author, with 1100+ citations.
Research training
Computational chemistry training informs scientific judgment and product decisions
- Shanghai University of Engineering ScienceMS, Materials Physics and Chemistry
- The University of AlabamaPhD, Chemical Engineering
- Northwestern UniversityPostdoc, Computational Chemistry
- DP TechnologyAI Product Manager (chemistry and materials)
| Research system | My research | Judgment developed |
|---|---|---|
| Catalytic reactions (MS) | Reaction pathways, transition states, and microkinetic modeling | Compare competing pathways and product distributions |
| Ionic liquids (PhD) | Ionic polarity and solvation affinity indices | Use interpretable metrics for materials screening |
| Circular polymers (postdoc) | Polymerization thermodynamics, polymerization and depolymerization mechanisms | Assess computational predictions in specific experimental systems |
Selected projects
From evaluation results to specific changes
Work at DP Technology spans four areas: the chemistry edition, agent development, scientific evaluation, and project proposals and delivery. All use the same process. Each card states the contribution: team product, development I led, or independent delivery.
Data quality
5-million-molecule quantum chemistry library Independent work
A 5-million-molecule quantum chemistry library, with QC built into production
Built quantum chemistry data for molecular representation learning. Independently completed the pipeline from source selection and computation to delivery. Added failure classification, monitoring, and automatic retries to production; delivered computation status with the data.
5,005,979molecular computation records
97.88%xTB convergence across all records
100%ORCA single-point convergence across all records
- UniMol-v2Select for scaffold diversity
- Normalize and deduplicateInChIKey
- GFN2-xTBGeometry optimization
- ORCAr²SCAN-3c single point
- Post-processing and QCFields and computation status
- ParquetDelivery
Monitoring, error reporting, failure retries
- Identifiers and structures
- 10
- xTB properties
- 6
- ORCA properties
- 9
- Derived properties
- 4
- QC and provenance
- 8
- Evaluation
- Convergence across all records reported separately for xTB and ORCA.
- Root cause
- Failed jobs recorded by type.
- Change
- Monitoring, error reporting, and automatic retries by failure type.
- Retest
- Retested failed jobs after retries. Separately, all 200 uniformly sampled API records passed consistency checks.
Bohrium Science Navigator, chemistry edition Evaluation design and diagnosis; implementation by the team
Substance search: distinguish the cause before changing the retrieval path
The same search failure can come from database coverage, retrieval, ranking, or result fields. Designed four evaluation layers to turn findings into change proposals, reproduction packages, and acceptance criteria.
- QueryName, Chinese name, SMILES, CAS
- RetrievalAre candidates found?
- RankingIs the right substance near the top?
- ResultsAre fields and pages sufficient?
- External benchmarkHow does coverage compare with common services? 1,690 deduplicated substances.
- In-library self-lookupCan indexed substances be retrieved? 500 records.
- Failure-stage diagnosisWhich stage failed?
- Manual inspectionDo page results fit research workflows? Not yet tested.
- Evaluation
- External benchmarking, in-library self-lookup, failure-stage diagnosis, and manual inspection each answer a different question.
- Root cause
- Ranking problems, retrieval problems, missing fields, and coverage gaps.
- Change
- Change proposals and reproduction packages.
- Retest
- Acceptance criteria defined; implementation by the team.
Other work on the chemistry edition My evaluation and validation; parser changes by the team
Turn Q&A, structure checks, and sample evaluation into materials for the next round of data changes
Generated 1,687 simulated chemistry questions from user profiles, providing questions and rules for Q&A review. Validated 3,006 structures across 10 patents and sent findings to developers by error type. Automated six-module evaluation with manual review produced 8 reports, one report per evaluation scope.
Routes and evidence
Retrosynthesis Agent Led product, orchestration, frontend, backend, and evaluation
Retrosynthesis Agent: three layers with explicit validation limits
Try online
Public version, no sign-in; jobs are visible only to you; limits apply per network. (Chinese interface)
Single-step models are open source. Led product and system development, with intern assistance on chemistry judgment. Separate layers handle candidate proposal and multi-step search, followed by checks and report writing.
Layer 1 Single-step candidate proposal
Reaction database searchReaction records
Neural modelsRootAligned + Graph2Edits
ORDOpen reaction records
Similarity searchECFP4
Merge and deduplicate candidates, then sort by consensus and rank
Repeated calls
Layer 2 Multi-step search
Best-first AND-OR search repeatedly calls the first layer and stops when leaf materials are purchasable. The inventory is ZINC in-stock building blocks plus a whitelist of common reagents.
Layer 3 Checks and report writing
- Parse input
- Propose candidates
- Forward reaction checkFirst route step only, for now
- Evidence gradingFirst route step only, for now
- Chemistry reviewFirst route step only, for now
- Deduplicate
- Write report
Retrosynthesis evaluation My evaluation design and offline comparison
Evaluation informs model changes and defines the limits of trust
L1–L3 have offline tests of different scope. At L4, the expert and experimental layer is still to come.
- L1 Literature alignmentCheck matches to literature routes
- L2 Like-for-like comparison160 marketed drugs and clinical candidate molecules
- L3 Chemical correctnessPurchasable starting materials, valid candidate structures, plausible reaction sites
- L4 Experts and experimentsEvaluate acceptance by chemistry experts; still to come
Offline comparison: first-disconnection diversity in the top three delivered routes
- Evaluation
- Offline comparison on 190 targets: whether the first disconnections differ across the top three delivered routes.
- Root cause
- The top three routes from the previous default often shared the same first disconnection.
- Change
- Changed the default to RootAligned + Graph2Edits.
- Retest
- Offline comparison on the same targets: the proportion rose from 27.4% to 37.9%.
In a single case, the same fixed 2-step route was held constant. After the Bohrium database was added, both steps gained reaction records, conditions, and citable sources. This shows evidence coverage only, not that the route is better.
Current limits: deeper steps have no forward reaction check or evidence grading yet; the expert and experimental layer is still to come.
Integrating retrosynthesis into ChemMaster as its initial capability. The local workbench connects research sessions, substance and literature search, source display, report export, and molecular drawing input.
Scientific tasks
SuperChem Led evaluation, root cause analysis, and ablation
SuperChem root cause analysis changed the product direction
On 500 retrieval-resistant chemistry multiple-choice questions, multiple frameworks, each paired with its own baseline, showed no statistically significant improvement. Root cause analysis and ablation located the bottleneck in model chemistry reasoning for this question set and the integration methods tested.
- Test integration methodsEach paired with its own baseline
- Root cause analysisFind the failure stage
- Ablation and reviewChemistry reasoning bottleneck
- Product directionTraining corpus for LLMs
- Evaluation
- Multiple frameworks, each paired with its own baseline, showed no statistically significant improvement.
- Root cause
- The bottleneck was in model chemistry reasoning for this question set and the integration methods tested.
- Change
- Changed the product direction from selling chemistry databases to providing training corpora for LLMs.
- Retest
- Later changes still need a new evaluation. Current findings apply only to this question set and the integration methods tested.
Trace Team collaboration; responsible for requirements, scientific judgment, and acceptance criteria
Trace: validate tasks, then calibrate difficulty
Two entry paths, one for automated task writing and one for screening and editing colleagues' tasks, share four checks. Package and deliver after multi-model runs and trajectory review. Feed customer feedback back into task-writing and screening strategies.
Automated task writing
Screen and edit colleagues' tasks
- EnvironmentStart from instructions
- Reference solutionActually computed in the same environment
- GraderCheck computed results
- DifficultyRepeated independent runs
- Multi-model runsTrajectory review
- Packaging and deliveryDeliver samples and trajectory data
- One LLM provider's medical VQA
- Delivered the first batch of trajectory data for 1663 questions; helped secure the collaboration.
- A national laboratory
- 20 sample tasks; prompted a price inquiry.
- Science benchmark project
- Demo delivery is under validation.
- Another LLM provider's RL task source
- Delivered 10 sample tasks.
- Evaluation
- Task validity, difficulty through repeated independent runs, and trajectory review.
- Root cause
- Used model runs to locate task difficulty problems.
- Change
- Iterated task-writing, screening, and delivery pipelines.
- Retest
- Multi-model runs; adjusted strategies using customer feedback.
Other projects
Product prototypes · business workflows My prototypes; team project
Product prototypes for a major project
About two months into the job, taught myself Vibe Coding and completed 246 of 458 third-level feature prototypes: 184 for knowledge graphs and 62 for deep mining. Project winning bid: over RMB 100 million.
Knowledge graphs: 184 featuresMining algorithms: 62 features
chem-lit2kb-pipeline My project · OCSR and validation
Chemistry literature extraction and validation pipeline
Extract structured knowledge from literature PDFs: molecules, reactions, tables, and spectra. Combine UniParser + MinerU parsing, cross-engine voting for optical chemical structure recognition (OCSR), G0–G4 five-level validation, and closed-loop correction of failed cases. Compare SMILES through RDKit InChIKey.
github.com/Xiaoyangliu-93 12 project repositories
Open-source code portfolio
Other tool repositories include an OCSR quality evaluator, patent LLM classification and cleaning tools, a multi-source scholar output aggregation pipeline, molecular gold-standard dataset tools, and a presentation-generation agent. All are independent work, organized by project as code-only repositories.
Long-term direction
Long-term direction: a forward reaction validator
This is the direction I want to pursue long term; there are no results yet.
Proofs or execution can check some claims in mathematics and code. Chemical reactions depend on real conditions and need experimental feedback. I want to work on a forward reaction validator: grade literature precedents, model predictions, mechanisms, and energy barriers by evidence strength, calibrate confidence, abstain when uncertain, and keep recalibrating against experimental results.
- Reliable data and routes
- Forward reaction validator
- Computation and experimental validation
- Property prediction and discovery
I start from 7 years in computational chemistry, using DFT and microkinetic modeling in long-term comparison with experimental groups, plus data quality, retrosynthesis, and evaluation experience from my current role.
Work and research experience
From computational chemistry research to AI4Science products
DP TechnologyBohrium Science Navigator
2025.08–present
Beijing
2025.08–present
Beijing
AI Product Manager (chemistry and materials)
Reporting to the head of product and data, a partner at the company
Scope: chemistry data products, high-throughput computation, agent development, scientific evaluation, and national and district project delivery.
1. Scientific data platform: Bohrium Science Navigator, chemistry edition (positioned against SciFinder / Reaxys)
- 2025.08–present. Search and parsing products for chemistry researchers. Responsible for data quality, core datasets, and external corpus delivery.
- 5-million-molecule library (independent work): 2026.01–04. Built quantum chemistry data for molecular representation learning. Selected UniMol-v2 for scaffold diversity and deduplicated by InChIKey. Used a coding agent to build a high-throughput pipeline for GFN2-xTB optimization and ORCA r²SCAN-3c single points, with automatic monitoring and retries. Delivered 5,005,979 records in 2026.05 with 37 fields: identifiers, 3D structures, global, atomic, and bond-order properties, and QC. Full-dataset xTB and ORCA convergence rates were 97.88% and 100%, respectively. View the computation pipeline
- Chemistry Q&A evaluation corpus: 2026.02–03. Generated 1,687 simulated questions from user profiles. Automatically collected Biomni Agent answers as a gold-standard test set; the gold labels are that agent's answers. Provided questions and rules for Q&A review.
- Database product definition and data specifications: Built gold-standard datasets. Used evaluation to guide product definitions, data dictionaries, schema design, and product page specifications for substance and reaction databases.
- Patent molecule validation: 2025.12–2026.02. Built a source-image comparison validation page and molecular structure editor (Chinese interface). Validated 3,006 structures across 10 patents. Returned the findings by error type, and developer colleagues revised the parsing code from them.
- External sample-data evaluation: 2026.05–06. Independently completed automated evaluation of six module types: PDF parsing, molecules, substances, reactions, image-text content, and VQA, with manual review. Produced 8 reports, one report per evaluation scope, and an Evaluation report collection (Chinese interface) for data iteration.
- Substance search: 2026.07–09. Designed four evaluation layers to compare our substance database with other common databases. Proposed changes and established a baseline for database quality iteration. View the four evaluation layers
2. Agent development: Retrosynthesis Agent and ChemMaster
- 2026.04–09. Led product definition, orchestration, frontend, backend, and evaluation. Participated in algorithm and evaluation research with intern-provided chemistry judgment. Integrating retrosynthesis into ChemMaster as its initial capability.
- Build and operation: Built a Retrosynthesis Agent with open-source single-step models and multi-step search frameworks. Seven modules cover structure parsing, route proposal, forward reaction check, evidence grading, review, deduplication, and report writing. Integrated best-first AND-OR search and connected tiers of commercially available compounds as the basis for deciding starting materials. Made the internal agent available for evaluation. The web page supports SMILES, Chinese names, drawings, and images; parses rules from a one-sentence route request; and generates a full synthesis route report. View the workflow
- Evidence and review: Designed stepwise evidence and forward reaction check displays, distinguishing strong reaction-level evidence from weaker substance-level evidence. The scoring covers feasibility, cost, evidence, safety, conditions, and purchasable starting materials. The service layer supports specifying the first bond break and required intermediates, and ranking by greenness, cost, and evidence.
- Evaluation system: Built L1–L4 evaluation (literature alignment / like-for-like comparison / chemical correctness / experts and experiments) and an internal iteration testbed. Compared four approaches: platform assembly, retrieval augmentation, general-purpose agents, and an in-house local build. The approaches differ by whether they include a forward reaction check and whether evidence is tied to the bond break. The Claude Science three-arm evaluation (bare model / plus public tools / plus the Bohrium database) examines how different agent architectures and databases affect retrosynthesis results.
- Extending ChemMaster: 2026.08–09. Built a local chemistry agent workbench connecting research sessions, substance and literature search, source display, report export, and molecular drawing input. Implemented three-layer collaboration for domain-specific proposals, cross-checking, and a synthesis report. Built a regression evaluation framework for SN 1687 questions.
3. Scientific evaluation: scientific agent task sources and benchmarks
- 2025.09–present. Trace, a source of scientific agent tasks, has been the current focus since 2026.07. Team collaboration; responsible for external requirements, scientific judgment, and receiving task sources and delivering sample tasks for two scientific agent task-source projects.
- Medical Q&A: 2025.09–10. Automated API evaluation on 325 USMLE questions, comparing Bohrium Science Navigator with ChatGPT, Gemini, and Claude. Classified errors to inform Q&A improvements for the medical edition.
- SuperChem evaluation: 2026.05–06. Systematically evaluated multiple frameworks on 500 retrieval-resistant chemistry multiple-choice questions across model architectures, including general tools, database tools, parallel agent injection, and native MCP. Each framework showed no statistically significant improvement against its own baseline. Root cause analysis and ablation identified model chemistry reasoning as the bottleneck. Changed the product direction from selling chemistry databases to providing training corpora for LLMs. View the decision process
- Scientific training corpus delivery for LLM providers: Built two automated delivery pipelines: one for writing tasks, the other for screening, editing, validating, and running tasks delivered by colleagues.
- One LLM provider's medical VQA: 2026.07–08. Built a question-screening, validation, and grading replay pipeline. Removed questions unanswerable without the image, ran multiple rounds of spot checks on the specified models, and made semantic grading replayable. Delivered the first batch of trajectory data for 1663 questions; helped secure the collaboration.
- National laboratory samples: 2026.09–10. Delivered 20 sample tasks; prompted a price inquiry.
- Science benchmark project delivery: Since 2026.09, responsible for task package screening, validation, review, and acceptance criteria. Task review and writing run in parallel; demo delivery is under validation.
- Another LLM provider's RL task source: Since 2026.10, responsible for task package screening, validation, review, and acceptance criteria. Delivered 10 sample tasks.
4. Project proposals and B2B / B2G delivery (project manager + core researcher)
- Task 1 of the 2030 National Science and Technology Major Project on AI-enabled organic chemistry: As a core member, wrote the midterm report, task outline, and test outline; completed the one-year review in 2026.06. As a core patent drafter, prepared the disclosure and figures, worked with the patent agent, and reviewed and revised application documents; the application has been accepted for processing. Task budget: RMB 15 million; project budget: RMB 75 million. Worked closely with the groups of Rong Zhu, Linfeng Zhang, Chen Zhu, and Jian Jiang at Peking University; Shanghai Jiao Tong University; Eastern Institute of Technology, Ningbo; and the Institute of Chemistry, Chinese Academy of Sciences, respectively.
- Haidian District open-competition project: 2025.10–2026.10. Participated in the proposal as a core member. The project uses the chemistry edition (interface partly in Chinese) as its platform, based on Bohrium's scholar, substance, and reaction databases and DP Technology tools including UniParser and UniMiner. Responsible for testing and acceptance materials; wrote the test outline and coordinated with CNAS institutions. Passed third-party technical testing. Consolidated requirements from 36 pilot applicants into a product iteration report. Project budget: RMB 40 million. Completed the formal acceptance review in 2026.10.
- Peking University AI4S corpus: 2025.08–10. Wrote application materials for mathematics, physics, and chemistry; helped the project win the bid. The project shares over RMB 1 billion in funding with the Chinese Academy of Sciences at 1:1.5.
- Guokexin major project prototypes: 2025.10. Two months into the job, taught myself Vibe Coding workflows and completed 246 of 458 third-level feature prototypes: 184 for knowledge graphs and 62 for deep mining. Created a semi-automated prototyping workflow for team reuse. Project winning bid: over RMB 100 million. Demos: Knowledge graphs (Chinese interface) · Mining algorithms (Chinese interface).
- Customer proposals: For research data and knowledge search, wrote a chemistry corpus proposal and quality test report for a university. Discussed chemistry and biology databases with another university and gave a product demonstration in English to a multinational company. Joined requirements discussions with a state-owned enterprise research institute and intellectual property organizations. Supported a patent classification project and provided SaaS and locally deployed RAG, knowledge base, and multimodal processing proposals as needed.
Northwestern University2022.08–2025.07
Postdoctoral researcher
Advisor: Linda J. Broadbelt (Member of the U.S. National Academy of Engineering)
The RePLACE project and the BOTTLE project
- Circular polymer design: Built a polymerization thermodynamics and ceiling temperature calculation workflow to assess monomer polymerization and recycling conditions. Evaluated ceiling temperature as a predictive design parameter. Published as first author in Cell Reports Physical Science (2024).
- Polymer depolymerization: With Tobin J. Marks's experimental group at Northwestern University, used DFT to study lanthanide-catalyzed depolymerization mechanisms of nylon-6 and other polymers. Compared reaction pathways to explain catalytic depolymerization. Published collaborative work as second author in Chem (2024) and Green Chem. (2025).
- Anionic ring-opening polymerization: With Eugene Y.-X. Chen's experimental group at Colorado State University, studied base-catalyzed ring-opening polymerization mechanisms of redesigned lactones and lactams to support recyclable nylon-6 variant design. Published collaborative work as second author in Angew. Chem. Int. Ed. (2024).
Collaboration within Northwestern University
- Siloxane exchange reactions: Studied associative siloxane exchange in covalent adaptable networks with amide-based internal catalysis, with John Torkelson's group.
- EPR calculations of radical intermediates: Used EPR calculations to study radical intermediates during fluoroethylene carbonate (FEC) reduction and help explain solid electrolyte interphase (SEI) formation, with Jeffrey Lopez's group.
- Team support: Provided simulation support to team members on industrial collaborations including Dow, supporting interdisciplinary work.
The University of Alabama2018.08–2022.07
PhD research assistant · Chemical Engineering
Advisor: C. Heath Turner
- Gas dissolution and separation: Proposed an ionic polarity index and Solvation Affinity Index for rapid ionic liquid screening. Used molecular simulation to analyze gas dissolution in multivalent ionic liquids. Worked with Jason E. Bara's group at The University of Alabama on carbon dioxide absorption solvents. Published related first-author work in J. Phys. Chem. B and Chem. Eng. Sci.
- Desalination: Used DFT to analyze fundamental interactions in solvent extraction. Worked with Steven T. Weinman's group at The University of Alabama and Ngai Yin Yip's group at Columbia University on computational and experimental studies of amine-based temperature swing solvent extraction. Published as first author in J. Mol. Liq. (2022). Received the Award for Outstanding Research by a Doctoral Student (2023, 1 recipient per year).
Shanghai University of Engineering Science2015.09–2018.05
MS research assistant · Materials Physics and Chemistry
Advisor: Wei An
- Biomass hydrodeoxygenation: With Daniel E. Resasco's group at the University of Oklahoma, used DFT to study m-cresol and guaiacol reaction pathways on NiFe and Fe-alloyed (Ni, Pt) surfaces. Developed transition-state scaling relationships and C–O bond-breaking descriptors, and predicted product distributions with microkinetic modeling. Published as first author in J. Catal. and Catal. Sci. Technol. (2018).
Skills
Where research, engineering, and product work meet
- Product and delivery: User research with scientists and requirement prioritization, PRDs and product prototypes, data dictionaries and acceptance criteria; B2B / B2G proposals, test outlines, and third-party test coordination.
- Scientific evaluation: Evaluation set and grading rule design, paired comparisons on the same tasks and significance testing, ablation and root cause analysis, and task difficulty calibration through repeated independent runs.
- Data and agent development: High-throughput computation and chemistry data pipelines (structure normalization, InChIKey deduplication, QC), agent orchestration and evaluation; Python, TypeScript / React, FastAPI, RDKit, Docker, Parquet. Develop with coding agents including Claude Code and Codex, with cross-review by different models.
- Computational chemistry: DFT (Gaussian, ORCA, VASP), semiempirical methods (xTB, CREST), molecular dynamics and Monte Carlo simulation (GROMACS, LAMMPS, Cassandra), and microkinetic modeling. Applied for and used XSEDE / ACCESS high-performance computing resources.
- Languages: Chinese and English (7 years of PhD and postdoctoral work in the U.S.).
Education
Academic background
The University of Alabama · PhD, Chemical Engineering
2018.08–2022.07
Shanghai University of Engineering Science · MS, Materials Physics and Chemistry
2015.09–2018.05
Jilin Normal University · BS, Applied Physics
2011.09–2015.06
Publications
Computational chemistry · polymer circularity and catalysis
30 journal papers: 11 as first author, 14 as second author, and 12 in collaboration with experimental groups. Selected papers below; full list on Google Scholar / ORCID.
30 Journal papers
11 First author
14 Second author
12 Experimental collaborations
1100+ Citations
Cell Rep. Phys. Sci. 2024 Liu, X.; Kozarekar, S.; et al. Understanding ceiling temperature as a predictive design parameter for circular polymers.
Angew. Chem. Int. Ed. 2024 Tian, J.-J.; Liu, X.; Ye, L.; et al. Redesigned Nylon 6 Variants with Enhanced Recyclability, Ductility, and Transparency.
Chem 2024 Ye, L.; Liu, X.; et al. Catalyst metal-ligand design for rapid, selective, and solventless depolymerization of Nylon-6 plastics.
J. Phys. Chem. B 2021 Liu, X.; et al. Gas Solubility in Multivalent Ionic Liquids: Insights from Molecular Simulation.
Full publication list: Google Scholar · ORCID. Gave 8 first-author oral presentations in total at AIChE annual meetings and FOMMS.
Honors and awards
Honors · computing resources
Received the American Institute of Chemists (AIC) Student Award (2022) and Graduate Council Fellowship, The University of Alabama (2021–2022, award rate below 1%). As applicant, received about 14.6 million SUs of ACCESS / XSEDE computing resources (estimated value about $64,700). Gave 8 first-author oral presentations in total at AIChE annual meetings and FOMMS.
Award for Outstanding Research by a Doctoral Student · The University of Alabama, 2023 (1 recipient per year)
2023