Skip to content
View johnquevedo's full-sized avatar

Block or report johnquevedo

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
johnquevedo/README.md

John Quevedo | Software Engineer

I'm a software engineer with hands-on experience building scalable backend systems and AI applications. I focus on backend engineering, distributed systems, and machine learning.

  • 🎓 Computer Science and Mathematics at Yale University
  • 💻 Interests: Backend Engineering, Distributed Systems, and Machine Learning
  • 🌱 Currently learning: System Design, Cloud-Native Architecture, and ML Deployment
  • 📫 Reach me: john.quevedo@yale.edu

Technical Skills

  • Languages: Python, Java, TypeScript, C++, Go, SQL
  • Frameworks & Tools: Spring Boot, React, Node.js, Express, PyTorch, Hugging Face, FastAPI, Docker, Git
  • Databases & Cloud: PostgreSQL, Redis, AWS (Lambda, Bedrock, CDK)
  • Other: REST APIs, WebSockets, RAG, MCP, Linux

📂 Featured Projects

Project Description
CI Failure Capture and Replay
  • Built a CI failure capture-and-replay pipeline that preserved code and runtime state, reproducing 50/50 controlled failures versus 0/50 clean reruns.
  • Engineered multi-tenant orchestration with idempotent jobs, lease-based queues, and checkpointed workers; preserved job state through duplicate webhooks, worker crashes, and database restarts.
  • Hardened replay containers with non-root, read-only, network-disabled execution and artifact verification; completed 172/172 controlled replays, including two 50-job concurrent runs.
Online Payment Repair
  • Built a data-repair workflow that traces corrupted payments to responsible code or configuration versions and generates hashed before-and-after repair plans.
  • Prevented 92 stale repairs from overwriting concurrent updates with optimistic locking while 8,787 writes completed without errors.
  • Repaired 100,000 records at 865 records per second with zero duplicates, incorrect fees, or ledger imbalances in controlled tests; completed all 5,000 repairs after injected failures.
Training Data Auditor
  • Built a local-first pipeline combining six label, similarity, and model-based signals into explainable review queues and auditable dataset patches.
  • Scaled exact-overlap profiling to 127,600 AG News rows in 9.8 seconds, surfacing 243 duplicate groups, 17 train/test overlaps, and 29 label conflicts.
  • Achieved 91.3% adjudicated precision across 100 blinded items reviewed by an independent non-domain reviewer; reproduced all 1,757 BANKING77 signals with alternate implementations.
CPU and GPU Inference Tuning
  • Automated hardware-specific inference tuning by benchmarking quantization, pruning, compilation, batching, and runtime choices against labeled data and detected device capabilities.
  • Implemented Pareto optimization under explicit quality-loss and p95-latency limits, exporting selected configurations as self-contained inference services with complete measurement manifests.
  • Reduced median batch-one latency by 79.6% and deployable artifact size by 21.0% with 0.0-point quality loss across three pretrained models on physical CPU and GPU targets.

📊 Github Highlights

  • Maintain 16 public repositories spanning open-source infrastructure, machine learning systems, academic work, and personal projects.
  • Shipped 4 production-oriented open-source systems with versioned releases, automated CI, reproducible benchmarks, containerized testing, and documented failure analysis.
  • Published reusable developer infrastructure through the GitHub Actions Marketplace.

📫 Connect with Me

Pinned Loading

  1. cpu-gpu-inference-tuning cpu-gpu-inference-tuning Public

    Evidence-first hardware-aware optimization across PyTorch, CUDA, ONNX, TensorRT, and Ray

    Python

  2. online-payment-repair online-payment-repair Public

    Production-safe correction of corrupted payment records with Java, Kafka, PostgreSQL, Spark, and Kubernetes

    Java

  3. ci-failure-replay ci-failure-replay Public

    Capture and replay pytest CI failures with content-addressed artifacts and isolated sandboxes.

    Python

  4. training-data-auditor training-data-auditor Public

    Explainable, local-first training-data quality and debugging for dataset maintainers

    Python