Resume / CV
Last updated 09/29/2026
Download my resume from the button on the right, or read a copy below! Visit the Publications / Talks and Projects pages for elaborations beyond what I can fit in my resume.
Leah Childers
Computational Scientist · Applied Mathematician · PhD in Mathematics (Mathematical Biology)
San Francisco, CA · leahchilders6745@gmail.com · LinkedIn · GitHub · Google Scholar · Portfolio
Computational scientist and applied mathematician who builds models of how antibiotic dosing drives bacterial killing and resistance. The standard measure of antibiotic potency, the MIC, assumes a steady drug level that patients never actually have. Created the Minimum Inhibitory Dose (MID), a dynamic alternative to the MIC for real dosing schedules, checked against static and dynamic in vitro experiments as well as population-level surveillance data: MID trends in U.S. and Norwegian gonorrhea surveillance data anticipate the CDC's 2020 ceftriaxone dose increase. The goal is dosing that clears infections without breeding resistance. Also writes fast scientific code in Python and Julia, has optimized a production ML model's inference, and trained a transformer language model from scratch. Seeking computational, research-scientist, and ML roles in biotech, pharma, and health.
SF Bay Area; PhD defended Sept 2026; available to start Q4 2026.
Research
Doctoral Researcher, Pennsylvania State University, Dept. of MathematicsAug 2022 – present
Thesis: Antibiotic Dosing and the Time-Dependence of Treatment: Mathematical Modeling Insights. Advisor: Dr. Jessica Conway; collaborator/co-advisor: Dr. Pia Abel zur Wiesch.
- The problem: antibiotic potency is measured by the MIC, a constant drug level that patients never actually have; real treatment is repeated doses that rise and decay. This work models what real dosing schedules do to bacterial killing and resistance, aiming for dosing that clears infections without breeding resistance.
- Minimum Inhibitory Dose (MID): a dose-and-schedule version of the MIC that gives the dose needed for a given dosing interval, or vice versa. In U.S. (GISP, 2016–2023) and Norwegian (2016–2025) gonorrhea surveillance data, MID trends anticipate the CDC's 2020 ceftriaxone dose increase.
- Built on Floquet theory: turns "does this regimen control the infection?" into a stability question about repeated doses, solved analytically where possible and numerically otherwise; one metric covers antibiotics usually split into "time-", "concentration-", and "exposure-dependent" categories
- Recovers the known frequent vs. once-daily schedules of five antibiotics; its ranking of treatment difficulty is consistent with independent hollow-fiber infection experiments
- Constant vs. periodic dosing (published): proved that, at the same total drug exposure (AUC), the shape of a drug's dose-response curve decides whether constant or periodic dosing leaves fewer bacteria, so "hit hard and early" may not be a universal rule for total bacterial reduction
- Drug combinations (MID isobole): given a kill target, finds every dose-and-schedule pairing of two antibiotics that reaches it. On published meropenem–tobramycin P. aeruginosa experiments, reproduced the observed ranking of all five tested regimens; both combinations were labeled "synergistic," but only one cleared the infection. Also predicted better-performing regimens.
- Resistance risk: stochastic model of whether resistant bacteria survive under real, fluctuating dosing (branching process, checked against 10k Monte Carlo runs per dose). The riskiest doses are those just high enough to hold the drug-sensitive strain in check (near its MID), and the schedule that clears bacteria fastest is not always the one that best prevents resistance.
- Clinical prediction: for an established, highly resistant gonorrhea throat infection, the models predict a single standard 1000 mg ceftriaxone dose can fail; a second daily dose cuts the resistant strain's survival chance from near 100% to about 7%, and a third to near zero
- Drug–target binding (receptor occupancy): extended the MID to a mechanistic model where drug effect depends on how many of each cell's targets are bound (184 equations for ciprofloxacin), then to two drugs hitting different targets. Built a Julia solver ~130× faster than full eigendecomposition. Shows cells cycling through binding states even while the total bacterial count holds steady, which a standard dose-response curve cannot capture.
- Scientific computing: built all simulation and analysis code from scratch in Python and Julia (stochastic PK/PD simulators, stiff ODE solvers, a matrix-Floquet stability solver); ran 50M+ simulations across hundreds of dosing regimens, with performance-critical Julia code ~10× faster than parallelized Numba. Code for the published paper and a ceftriaxone/N. gonorrhoeae PD database are public; remaining code goes open source as each paper is submitted.
Publications
- L. Childers, P. Abel zur Wiesch, J. M. Conway. "A General Analytic Approach to Predicting the Best Antibiotic Dosing Regimen." Journal of Biological Dynamics (2026). doi:10.1080/17513758.2026.2641302
- L. Childers, S. Abel, B. Børud, P. Abel zur Wiesch, J. M. Conway. "The Minimum Inhibitory Dose (MID): A Dynamic Alternative to the Minimum Inhibitory Concentration (MIC) for Guiding Antibiotic Treatment." Soon to be submitted.
- L. Childers, P. Abel zur Wiesch, J. M. Conway. "The Minimum Inhibitory Dose (MID) Isobole: A Framework for Selecting Antibiotic Combination Doses and Schedules." In preparation.
- L. Childers, J. M. Conway. "Thresholding Resistance Risk under Time-Varying Antibiotic Dosing." In preparation.
- L. Childers. Antibiotic Dosing and the Time-Dependence of Treatment: Mathematical Modeling Insights. Ph.D. dissertation, Pennsylvania State University (2026).
Experience
- Contract ML engineering, San Francisco music-AI startup (Jul – Aug 2026): paid independent contractor; inference optimization for a production transformer model (including speculative decoding); work merged and deployed to production. Details under NDA.
- Instructor of Record, College Algebra, Penn State (Jan 2023 – Aug 2025): taught 50–100 students/semester with full course ownership: wrote and delivered lectures, wrote and graded exams, held office hours.
- Coordinator Assistant, Math 021, Penn State (Aug 2024 – May 2025): wrote homework, worksheets, and exams; ran biweekly instructor meetings.
- Technical Writer, Beam.cloud (YC W22) (Mar – May 2025): wrote developer-facing articles on training, developing, and selecting ML models (e.g., LLM Parameters for Developers).
Selected Projects
- NeurAllegro (2024 – present): a language model that reads and writes sheet music as MusicXML. Designed a 632-token tokenizer that keeps the dynamics, articulations, and slurs MIDI discards; blind-benchmarked against 15 other formats (notation-fidelity F1 0.95 vs. 0.84 next best). Built a 413M-token training corpus from 53,000 public-domain piano scores with leakage-free train/val/test splits and a custom music21 fork, then trained a 25.5M-parameter transformer from scratch in PyTorch on a single laptop GPU. On its first training run it writes complete, valid piano scores with multi-voice harmony, cadences, and full phrases, and given the opening of a piece it has never seen, continues it in the same style.
- LearnFlow (2025): designed and deployed a full-stack learning-management platform (Next.js/React, Supabase/Postgres, OpenAI & Gemini APIs, Vercel) introducing recursive "flows" for course structure and flexible non-numeric grading. Cut course-build time ~90% (measured by rebuilding a real Penn State course in Canvas vs. LearnFlow); designed with Penn State faculty input.
Skills
- Modeling & methods: PK/PD and dose–response modeling, dosing-schedule optimization, drug–target binding (receptor occupancy), stochastic simulation & branching processes, Monte Carlo, stiff ODEs, Floquet theory, numerical linear algebra
- Languages: Python, Julia, R, MATLAB, SQL (PostgreSQL/SQLite)
- Scientific computing & ML: NumPy, SciPy, JAX, Numba, Julia ecosystem, PyTorch (transformer training from scratch, FlexAttention), scikit-learn; ML inference optimization (profiling, speculative decoding)
- AI / LLM tooling: LLM CLIs/APIs/agents
- Data viz: matplotlib, Plotly, Dash, ggplot2
- Dev & web: Linux, Git, Bash, LaTeX; Next.js, React, TypeScript, Supabase, Vercel
Education
- Ph.D. in Mathematics, Pennsylvania State University: defended Sept 2026
- B.S. in Mathematics, Virginia Tech (2022): summa cum laude; completed in 3 years
- Relevant graduate coursework: PK/PD Modeling, Bayesian Statistics (hierarchical models in RStan), Numerical PDEs, Regression Models, Deep Learning, Epidemic Modeling
Presentations
- Contributed talk, "The Minimum Inhibitory Dose (MID): A Dynamic Analogue to the Minimum Inhibitory Concentration (MIC) for Periodically Driven Antibiotic Treatments," SIAM Conference on the Life Sciences, July 2026.
- Organizer & invited speaker, minisymposium "Mathematical Modelling for Infectious Diseases and Interventions," SIAM NNP Section Conference, Nov 2025. Talk: "A General Analytic Approach to Predicting the Best Antibiotic Dosing Regimen."
- Dissertation defense, Penn State, Sept 2026.
- Invited talks: Virginia Tech MathBio seminar (date TBD), Penn State Math Club (Spring 2025).
- Poster, Penn State Graduate Student Exhibition (Spring 2025).
Honors & Awards
- August and Ruth Homeyer Graduate Fellowship, Penn State (2022)
- SURE ("Speakers and Undergraduate Research Engagement") Fellowship, Virginia Tech (2022)
- Ray A. Gaskins Scholarship in Math, Virginia Tech (2021–2022)
- T. W. Hatcher Math Scholarship, Virginia Tech (2021–2022)
- Jessie S. Yee Memorial Prize, 1st place, Virginia Tech School of Performing Arts (2022), for the orchestral work "Symphony No. 1: The Myth of Er" (announcement)