Taylor Johnson
Software Engineer · AI Systems & Developer Tools
New Albany, Ohio
johnsontaylor6320@gmail.com | +1 220-246-8386
https://becker63.digital/ | https://github.com/becker63

Software engineer focused on making complex systems understandable and efficient through explicit representations, typed contracts, and reproducible tests. My work connects real-time grid state coordination, production HPC authentication, and AI evaluation and execution infrastructure.

TECHNICAL DEMOS & RECOGNITION
AI Tinkerers Columbus — Technical Demo Presenter | June 2026: Presented SearchBench; featured as a Community Favorite in Top AI Demos #31.
Demo: https://aitinkerers.org/talks/rsvp_t4IMPq7lJnc
Roundup: https://post-training.aitinkerers.org/p/top-ai-demos-31

INDEPENDENT ENGINEERING & RESEARCH
Attune — Independent engineering & research project
Source: https://github.com/becker63/attune
Interactive explanation: https://atlas-live-five.vercel.app/
- Built a typed repository-analysis instrument in Flix with Grit extraction and bounded Datalog queries; separated semantic policy from narrow Java/native interfaces and pinned the Bazel/Nix build.
- Designed reproducible localization and review evaluations with gold-free prediction inputs, frozen receipts, exact memoization, and shared Bazel artifacts; integrated BuildBuddy remote caching and execution.
- Measured scheduling and replay costs: historical AttuneFlix offline evaluation preserved results across 61 JS/TS cases while reducing 4,301 seconds of sequential shard work to 51.7 seconds of cold remote wall time; exact replay made no new provider calls.
- Built an installable agent-tool contract pipeline using DeepSeek Harness and Typert, rejecting unsupported schemas; validated transport/tool boundaries alongside semantic laws, renderer checks, and six README examples regenerated from real extraction.

SearchBench — Coding-agent search evaluation harness
Source: https://github.com/becker63/searchbench-go
- Built a Go/Pkl experiment surface and Python evaluation harness to compare coding-agent search strategies using typed tasks, scoring models, backend adapters, and reproducible run bundles.
- Recorded search quality, token usage, cost, and classified failures so candidate strategies can be compared against a baseline with inspectable evidence.

PROFESSIONAL EXPERIENCE
Ohio Supercomputer Center — Security & Infrastructure Documentation Intern | October 2024–November 2025
- Investigated and documented a production HPC platform, tracing OIDC, JWT, and LDAP identity flows through Apache modules, Nginx per-user processes, and Lua authentication hooks.
- Built a Docker reproduction of the authentication stack to debug behavior without privileged production access; investigated configuration layers and security-sensitive URL handling.
- Turned authentication and infrastructure findings into reproducible debugging procedures and documentation for onboarding and platform maintenance.

Tarigma Corporation — Frontend Engineer | 2021–2023
- Built React/Recoil state coordination for real-time electric-grid incident analysis, synchronizing multiple relay data feeds for reliability-sensitive operator interfaces.
- Extended Chart.js with draggable temporal annotations and implemented COMTRADE relay-data processing for electrical disturbance analysis.
- Integrated dynamic windowed diagnostic layouts with react-mosaic, keeping timing, state consistency, and debugging clarity central to the interface.

STEMTree — Programming Instructor | 2021–2023
- Taught Python, JavaScript, React, and computer science concepts to students ages 10–16; helped students debug programs and development environments through hands-on instruction.

SELECTED SYSTEMS & AI EXPERIMENTS
Local Inference Experiments: Modified llama.cpp’s Qwen embedding path on Apple M2/Asahi Linux with Vulkan. Removed unnecessary language-model-head work, preserving bit-identical vectors in the tested comparison; throughput improved from ~72.2 to ~78.8 tokens/s (~9%).
AI Program Optimization: Investigated DSPy Flex/GEPA executable-program optimization and compiler-feedback validation; implemented compatibility patches for recursive tool schemas and NumPy lazy imports. Kept candidate selection separate from evaluation.
Static Control Plane: Modeled Crossplane/KCL infrastructure with Nix-pinned schema generation, GitOps resources, and validation before deployment.
NixOS Infrastructure: Built declarative multi-node NixOS deployments with provisioning, Tailscale, distributed builds, and binary caching.
nftables Structure Fuzzer: Built structured Netlink mutation with Nim/C bindings, sanitizers, and reproducible Nix/MicroVM tests.

ADDITIONAL RECOGNITION
PicoCTF, Carnegie Mellon University | 2024: Placed 57th of 1,329 teams.
SkillsUSA Cybersecurity: Second place, state competition.
US Cyber Challenge Cyber Camp | 2025: Selected for practical security training and CTF exercises.

TECHNICAL SKILLS
Languages: Go, Python, JavaScript/TypeScript, Flix, Java, Nim, Bash
Systems & tooling: Linux, Nix/NixOS, Bazel, BuildBuddy, Buck2, Docker, Git, CI/CD, Kubernetes, Crossplane, FluxCD
Interfaces & evaluation: React, Recoil, Chart.js, Grit, Datalog, typed experiment models, metrics, reproducible artifacts
Testing & contracts: pytest, Hypothesis, BasedPyright, Ruff, LibCST/Fixit, compiler feedback
Infrastructure & security: OIDC/JWT/LDAP, Linux networking, fuzzing, Prometheus/Grafana

EDUCATION & CERTIFICATIONS
Computer science coursework — Ohio State University, Newark
Cybersecurity program — Eastland-Fairfield Career & Technical Schools
CompTIA Security+ | CompTIA Network+ | TestOut Network Pro