Nathan Landau
Staff ML Software Engineer
About
ML Software Engineer with a background in Mechanical Engineering, specializing in AI systems for physical industries: knowledge graphs, agentic LLM workflows, industrial data pipelines, and enterprise integrations. Experienced in Python, Java, and Node.js, with a focus on creating practical, well-integrated software that solves real-world physical and financial problems.
Experience
Staff ML Software Engineer
Oct 2026 - PresentDigital Iron builds an AI platform for heavy equipment dealerships, bringing together service data, machine history, and documentation to speed up troubleshooting, parts selection, and quoting.
- Work across the integration architecture and ML infrastructure behind Digital Iron's AI platform for heavy equipment service and parts.
- Contribute to the data ingestion pipelines that extract content from PDFs and load it into Meilisearch search indexes and structured data stores for the platform to query.
- Build the data pipelines, knowledge graph, and LLM agent systems that help dealerships diagnose equipment and source the right parts.
- Bring a mechanical engineering background to heavy equipment service, repair, and parts data.
Energy Model Developer
Feb 2025 - Sep 2026Ekotrope builds software for the residential construction industry to help raters and builders analyze energy efficiency and meet building codes.
- Started and co-built Automated Takeoffs, an LLM service that reads building-plan PDFs and extracts window data so raters no longer enter it by hand. Benchmarked Gemini, ChatGPT, and Claude, designed a multi-stage prompt chain that reached 88.9% extraction accuracy on about 200 plans at about $0.13 per plan, wrote a 46-threat security threat model, and shipped the RATER integration as a live beta in August 2026.
- Led the engineering side of Ekotrope's AI Taskforce, defining the company's policy and tooling for AI-assisted development, including a 5-phase human-led workflow, a Rule of Two safety posture that closes the untrusted-content leg of the lethal trifecta in config, and four Claude Code enforcement hooks at session start, pre-edit, post-edit, and pre-commit.
- Worked on the Java rating engine behind RATER and CODE, including energy code checkers, load and infiltration algorithms, and reports, on a GWT, Apache Tomcat, Maven, and Cog stack, and covered a weekly customer support shift.
- Designed and deployed a shared AI-assisted-development setup for the Java codebase, with 6 specialist subagents and auto-routing rules, 9 project skills, a risk-tiered guard system matching friction to blast radius, path-scoped rules, pre-commit gates, and configured MCP servers, then taught it in a Claude Code workshop for all developers.
- Built an MCP server and agent loop at Ekotrope's 2026 AI Hackathon that searches for the least-cost set of changes to bring a house plan into code compliance, checked against the Ekotrope energy engine.
- Implemented seven state and local energy codes, led the Florida 2023 code by mapping it against IECC 2021 section by section into the plan the team built from, and shipped ENERGY STAR v3.3 and ANSI/RESNET/ICC 301-2022 changes in the V5.1 and V5.2 rating-engine releases.
- Co-built Ekotrope CODE, a standalone Total UA compliance calculator for home builders, from reusable component tables to a live compliance panel and reports, and aligned its R-value and air-film assumptions with ASHRAE 90.1 Appendix A so ICC would audit the calculation.
Software Engineer (Contract)
Dec 2024 - Dec 2024Sekond is a financial data platform that brings transparency to alternative investments by organizing fragmented regulatory filings into real-time data for the Evergreen fund market.
- Developed Python scrapers for SEC filings to collect data on the holdings of publicly traded private equity funds.
- Architected automated data collection pipelines to keep the platform consistently updated with the latest financial data.
Software & Mechanical Engineer
Sep 2019 - Nov 2024Divert eliminates food waste by using IoT, automation, and integrated software platforms to recover and repurpose wasted food through nationwide infrastructure.
- Led migration of cloud infrastructure from AWS to GCP, consolidating environments and streamlining the engineering team's deployment workflow.
- Designed and implemented a Node.js API to expose SQL data and perform real-time calculations, and built the React frontend for the facility-monitoring dashboards it powered.
- Engineered a scalable computer-vision prototype on Google AI Studio, Cloud Functions, and Cloud Storage to identify food groups in grocery store waste streams.
- Automated ETL workflows using Python to extract and standardize order data from complex Excel files into centralized databases.
- Developed Python automation scripts to streamline weekly and monthly reporting, significantly reducing manual effort for the Customer Success team.
- Built the real-time monitoring system for the cold grocery supply chain: SciPy and TensorFlow algorithms that predicted each IoT sensor's position through the chain from its sensor data, surfaced in interactive Ignition SCADA dashboards across retail display cases and cold storage.
Systems Engineer
Jan 2019 - Jun 2019Casne Engineering is a consulting firm that provides engineering and technology integration for industrial and power applications.
- Designed and deployed Ignition SCADA interfaces for real-time power monitoring in mission-critical data centers.
- Established robust data integration pipelines between Ignition SCADA and OSIsoft PI Systems.
Research And Development Intern
Dec 2016 - Jan 2019Smith Engineering provides specialized engineering and energy consulting services focused on district energy systems, central plant optimization, and building energy efficiency.
- Developed machine learning models in Azure to forecast energy demand and predict utility bills based on known rates with 97% accuracy.
- Built the 2nd generation real-time energy dashboard for 41 Cooper Square using OSIsoft PI Vision, integrating BACnet/IP data for carbon emission tracking.
- Contributed to the 'Roadmap to 40 X 30' initiative for The Cooper Union, implementing energy management strategies that achieved significant kWh savings and Local Law 97 compliance.
- Led a team of interns to build Pi Vision dashboards for monitoring over 1,000 data points.
Technical Projects & Infrastructure
Autonomous Agent Company (Paperclip)
Self-hosted Paperclip company of 12 AI agents that takes Linear issues from intake through engineering, code review, release, deployment, and verification, with every agent reaching its models through a shared LiteLLM gateway.
- Built a 12-agent pipeline where a lead agent takes Linear issues and hands them to engineers, reviewers, a release agent that merges, a separate deploy agent, and a verifier before the lead closes the issue.
- Routed every agent through LiteLLM with its own virtual key, so no provider credential is stored in Paperclip and each agent can run on local or hosted models.
- Applied the Rule of Two to the pipeline and narrowed each leg: separate merge and deploy identities with separate keys, a sandbox whose network egress is fenced and verified by packet tests, and a policy check every 30 minutes that flags changes to agent models, grants, keys, or instructions.
Home Automation & Touch Panel
View on GitHub →Custom ESP32-S3 touch panel for Home Assistant, with open-source firmware that extends the Waveshare reference design to auto-discover and control lighting, climate, media, and scenes.
- Extended the Waveshare ESP32-S3 demo firmware to integrate with Home Assistant, adding entity auto-discovery, NVS caching for instant reboot, and an async command queue with 120ms coalescing.
- Added a WebSocket event subscription alongside REST polling, offline detection with auto-reconnect, and a Windows LVGL simulator for iterating on layouts without flashing hardware.
- Brought up the Waveshare ESP32-S3-Touch-LCD-7 (800x480 RGB, GT911 capacitive touch over I2C), configuring tear-free double-buffered PSRAM rendering for LVGL.
Proxmox Homelab Cluster
Hyper-converged 4-node Proxmox cluster managing 6 ZFS pools, automated snapshots, and multi-tiered disaster recovery across about 180 TB of raw storage.
- Resolved Docker-to-LXC routing conflict by engineering a custom systemd service that dynamically re-injects iptables FORWARD chain rules after Docker startup.
- Consolidated 20+ services into Docker-in-LXC containers on an isolated services VLAN and kept their original ports stable via transparent iptables DNAT forwarding.
- Engineered a 4-node quorum cluster with ZFS storage, a dedicated 10Gb storage fabric, and automated Sanoid snapshot retention policies.
- Designed 3-2-1 disaster recovery: tiered Sanoid snapshots, nightly deduplicated Proxmox Backup Server backups to a second node, and an encrypted rclone mirror of the backup store to Backblaze B2.
Self-Hosted Infrastructure
40+ applications including Immich, local LLM hosting, and media servers, deployed across a tiered Docker plus LXC topology with network isolation and zero-trust access.
- Orchestrated 40+ self-hosted applications across Docker and LXC containers, with GPU-accelerated local LLMs on Intel Arc and NVIDIA (VFIO passthrough), both routed through a unified LiteLLM proxy.
- Secured the infrastructure using UniFi VLAN isolation, a Cloudflare Tunnel into a Caddy reverse proxy with automated TLS, Proxmox firewalls, and Tailscale for zero-trust remote access.