Most engineering teams hit a wall when production reliability and HPC/research-grade infrastructure need to coexist. Web-scale distributed systems engineers don't speak HPC. HPC researchers don't ship production AWS code. Healthcare AI teams need both, and almost no one has both.

I've watched the same pattern repeat:

• Legacy systems quietly cap throughput, but nobody redesigns the data path — they tune it
• Research code lives on a laptop until it dies on the cluster
• "We'll add compliance later" becomes "we shipped, now what?"

The cost? 50% over-spending, 30% latency you can't fix with caching, and an LLM you can't actually ship into a regulated environment.

The rare engineer who bridges production AWS at scale, HPC-based ML/AI research, and regulated healthcare LLM deployment — three intersecting depths that almost never coexist in one person.

🏥 Regulated Healthcare LLM (funded RA, Summer 2026 →):

• HIPAA-oriented clinical LLM chatbot for MCI patient support at SLU HPC Lab — working prototype deployed (Docker + AWS ECR, TLS): device pairing, onboarding, AI-initiated proactive conversation, hands-free voice, coordinator dashboard, 690+ automated tests — Read More
• Server-side crisis detection (Code Blue escalation, 988 lockout) + prompt-level safety guardrails for a vulnerable population
• BAA-grounded data architecture (engagement metadata only, no raw audio/transcripts) feeding a future NIH R01 application

🔬 HPC + ML Research at SLU HPC Lab (M.S. CS, Dec 2026):

• Federated Learning parallelization with empirical Amdahl analysis (1.27× stable speedup, ~24% parallel fraction ceiling proven); now extending to multi-node FL + FedAsync on the SLU Libra cluster
• Gaussian Process Regression + LSTM pipeline for clinical force prediction with uncertainty estimation, on NVIDIA L40S under SLURM
• CARLA distributed autonomous-driving data collection with GIST (2 nodes × 8 A100, SLURM + Singularity) — contributed the watchdog + auto-resubmission job chain; targeting an SC2026 workshop paper

☁️ Production AWS at scale (4 years at Doverunner):

• Redesigned a synchronous-RDB watermark token system into a Lambda + Redis + SNS architecture → 5M+ race-free tokens across Seoul, Oregon, Frankfurt with 30% peak-latency reduction
• Replaced a Hybrik-based encoding SaaS with a self-hosted AWS EKS pipeline → 50% cost cut, 30% throughput lift
• Reduced 100-user video conference room initialization from 0.6s to 0.1s — a 6× user-facing latency improvement at TmaxWAPL

Mid-level (SDE II / SWE II) roles in:

• ☁️ Cloud / Distributed Systems Infrastructure
• 🤖 ML / AI Platform Engineering
• 🎬 Video / Streaming Infrastructure
• 🏥 Healthcare AI

📍 St. Louis, MO | Open to relocation anywhere in the U.S. | F-1 OPT + 3-year STEM extension eligible | Available January 2027

→ Reach me via LinkedIn or email tofoth@gmail.com.

• Funded RA project at SLU HPC Lab (advisor: Prof. Ted Ahn) — an iPad-based conversational AI for patients with Mild Cognitive Impairment (MCI), designed for a HIPAA-compliant clinical deployment as a feasibility pilot supporting a future NIH R01 application. Code private (clinical constraints) — architecture, safety design, and voice pipeline documented in the public showcase repo
• Working prototype deployed end-to-end (Docker Compose + AWS ECR behind TLS/nginx, CI/CD via GitHub Actions + SSM): React/TypeScript PWA kiosk, FastAPI backend, PostgreSQL with de-identified metadata only — covered by 690+ automated tests (pytest + Vitest, E2E smoke in CI)
• Safety enforced server-side: crisis detection with Code Blue escalation and a 988-crisis lockout flow before any LLM call, plus prompt-level guardrails. Data architecture captures engagement metadata only — no raw audio or full conversation transcripts
• Hands-free voice implemented: turn-based, server-relayed via AWS Transcribe / Polly with latency engineering — per-sentence streaming TTS, speculative early STT, prompt caching. Pilot: ~15 participants, 6 months
  Stack: FastAPI, React + TypeScript, PostgreSQL, Claude API, prompt engineering + safety guardrails, AWS (Transcribe, Polly, ECR), Docker, pytest + Vitest

• Run CARLA simulation workloads across 2 nodes × 8 NVIDIA A100 GPUs, scheduled with SLURM and containerized with Singularity, collecting data for the LAV and InterFuser autonomous-driving pipelines
• Contributed the watchdog + automatic-resubmission job chain so long-running collection jobs survive transient failures without manual intervention
• Targeting an SC 2026 workshop paper
  Stack: CARLA, SLURM, Singularity, NVIDIA A100, Python

• Identified the single-process bottleneck capping FL research throughput; prototyped Python ProcessPoolExecutor parallelization and diagnosed a PyTorch tensor-pickling deadlock at the IPC boundary, resolving it via NumPy serialization
• Measured ~1.27× stable speedup across 50–200 Non-IID MNIST clients on an 8-core CPU; proved an Amdahl ceiling (~24% parallel fraction), pointing the next iteration to multi-node FL rather than further single-machine tuning
• In progress: scaling FL to multi-node SLU Libra HPC and Asynchronous FedAvg (FedAsync-style); planned: CIFAR-10 + ResNet-18 workloads
  Stack: Python, PyTorch, NumPy, multiprocessing, MNIST, FedAvg

• GPR for thickness extrapolation: Found XGBoost collapsed to flat lines outside the training range (0.25/0.5mm), useless for predicting forces (Fx–Fz, Tx–Tz) on U6/U7 molars at unseen thicknesses (0.75/1.0/1.25mm). Re-architected with Gaussian Process Regression (Matern kernel) for μ ± 2σ uncertainty; augmented training with weighted-delta-method synthetic 0.75mm data to distinguish 1.0 vs 1.25mm predictions
• LSTM LOO + Partial Observation forecasting: Reframed as "given a new patient's early measurements, forecast later force/moment trajectory." Designed a 60-experiment matrix (5 cohorts × 3 horizons × 4 sheets) — train on 4 other cohorts plus target cohort's first (11−k) points, predict the last k ∈ {1,2,3} points
• Model: LSTM(128, 2 layers, dropout 0.2) → Dense(64) → scalar, 300 epochs with StepLR
• Complementary design: GPR covers spatial extrapolation (thickness), LSTM covers temporal extrapolation (14-day window). Trained on NVIDIA L40S GPUs under SLURM on the SLU Libra HPC cluster
  Stack: Python, PyTorch, GPyTorch, XGBoost, scikit-learn, pandas, numpy, SLURM

Redesigned the legacy synchronous-RDB token service into a non-linear-index Lambda + Redis + SNS event-driven architecture, to fix peak-time latency spikes and regional index collisions blocking scale-out, delivering 30% peak-latency reduction and 5M+ race-free tokens across Seoul, Oregon, and Frankfurt.

Stack: Kotlin, AWS Lambda, Redis, SNS, CloudWatch, RDS

Designed and built a containerized encoding pipeline on AWS EKS with GOP-aware video partitioning and Redis-based job control with auto-failover, to replace a Hybrik-based external SaaS that was cost-prohibitive and inelastic at scale, cutting encoding cost 50% and lifting throughput 30%.

Stack: Kotlin, Spring Boot, AWS EKS, Docker, Redis, FFmpeg

• M.S. in Computer Science (in progress) — Saint Louis University, USA, Aug 2025 – Dec 2026 (expected)
  Advisor: Prof. Ted Ahn | Lab: High Performance Computing Lab | GPA: 3.63 / 4.0
• B.S. in Computer Engineering — Ajou University, South Korea, 2015–2018