work.jackmann.xyz
← Home

All projects

Software, physics research, and science education.

  • 2026

    Beamline for Schools (BL4S)

    Project Lead

    Led a 7-person international team submitting a CERN Beamline for Schools proposal investigating biopolymer-based radiation shielding for Martian applications. Developed Monte Carlo simulations in Geant4, designed experimental protocols to CERN standards, and coordinated cross-timezone research workflows.

    • Project Lead for a 7-person international team
    • Geant4 Monte Carlo simulations for shielding effectiveness in particle beamline environments
    • Experimental protocols designed to CERN high-energy physics research standards
    • Biopolymer shielding evaluated for long-duration Mars surface missions
  • 2026

    OpenLab Experiment Workshop

    Project Lead & Founder

    Initiated and led OpenLab 2026 under the Youth Ambassadors East Asia & Pacific program. Secured U.S. State Department funding, managed budget and logistics, coordinated with embassy partners, and designed a workshop where participants conduct and test real scientific experiments.

    • Secured U.S. State Department grant through YA EAP program
    • Managed budget, logistics, and embassy coordination end-to-end
    • Designed hands-on experiment sessions where participants test scientific methods firsthand
    • Youth Ambassador leadership for East Asia & Pacific outreach
  • 2026

    Sawtooth Memory

    Author & Maintainer

    A high-performance memory framework that eliminates main-thread blocking during conversation compression. Messages ingest instantly to L1 while a background worker summarizes to L2 — critical facts are extracted into an immutable entity ledger (L1.5) via deterministic NER before the LLM can hallucinate them away.

    • Zero-latency ingestion — messages append instantly, compression runs in background
    • Entity Ledger (L1.5) with deterministic NER — 100% UUID and ID recall
    • 11.3× faster main-thread execution vs standard summary memory (benchmarked)
  • 2026

    AlphaRag

    Author & Maintainer

    A Chainlit-powered document assistant using parent-child chunking, hybrid dense+sparse search, and automated LLM metadata extraction. Each upload session gets an isolated in-memory vector store — multi-user safe with no disk persistence.

    • Session-scoped in-memory Qdrant with UUID isolation per upload
    • Parent-child retrieval — child ANN search, parent context expansion
    • Automated metadata extraction (title, author, date) via Groq + Pydantic
  • 2025

    Fact Hub Workflow Automation

    Developer

    Architected a server-side automation framework using Google Apps Script with a reactive state machine for Fact Hub Myanmar's content lifecycle. Integrated Google Workspace with Slack and email APIs, achieving 99%+ uptime and full data integrity across production environments.

    • Reactive state machine for end-to-end content lifecycle management
    • Google Workspace integrated with Slack and email notification APIs
    • 99%+ uptime with full data integrity in production
  • 2026

    Freelancer OS

    Full-Stack Developer

    A data-driven dashboard unifying time tracking, project management, and invoicing for independent professionals. Executive overview surfaces outstanding revenue, active projects, and lifetime billable hours — eliminating context switching across disconnected tools.

    • Executive overview with revenue, projects, and billable hours at a glance
    • Precision time tracking with fractional-hour logging (no rounding drift)
    • Playwright E2E CI/CD pipeline for auth and database critical paths
  • 2026

    AI Pipeline Toolkit

    Author & Maintainer

    An orchestration toolkit for multi-step AI workflows — webhook triggers, local LLM inference via Ollama, and LangGraph state machines for complex agent pipelines. Designed for teams that need control without running a full MLOps stack.

    • LangGraph state machines for multi-agent coordination
    • FastAPI webhook ingress with async job processing
    • Local inference via Ollama with fallback to cloud providers