David N. Juboor

I'm a hyper-curious ML engineer who keeps getting pulled into hard problems across fields I didn't start in. Pharma, finance, legal, entertainment, robotics, research — often more than one at a time.

Experience
··
  • ML lead for JARVIS — Imagineering's agentic AI platform, >80% adoption (~10k DAU)
  • Self-hosted the model stack — 16 H100s, fine-tuned open models, vLLM + LiteLLM
  • Invented self-maintaining agent tool routing + multimodal MoE — patents below
  • Multi-agent architectures and MCP integrations across hundreds of tools
  • Team of 6; moved frontier-model workloads to self-hosted inference — ~10x cheaper per run
  • Cross-org ML strategy across Imagineering, Disney Research, and Studios
  • Founding ML engineer for PhotoPass — 0 → 15+ products, $16M/yr in ML-driven revenue
  • Automated character photography across every US and EU park — 20M+ images/day
  • Anticipatory capture (15 → 60 FPS), group recognition, auto-edit, moderation
  • Grew the function from 1 to a 20-person org
  • Photographers displaced by the automation became its 40-person annotation team
  • Golden datasets behind Web Slingers: A Spider-Man Adventure
  • Core developer on Lexis+, the flagship legal research platform
  • Productionalized custom OCR and information extraction over 300M+ documents
  • Production pipelines for training, fine-tuning, and deployment of GPT-based foundation models
  • Led ML on AIVA — mortgage-workflow AI for top-10 banks (AWS re:Invent 2022)
  • Grew the applied deep learning team 1 → 12
  • Took production performance from 41% to 92%
  • Reduced overhead costs 30x via serverless architectures
  • Increased NLP transformer performance via novel pruning and ensemble methods
  • NLP unsupervised clustering for mission-critical financial reporting
  • Supported web-scraping infrastructure and DL recommendation systems
  • Deep learning anomaly detection for autonomous laboratory robotics
  • Reduced drug compound analysis resources 500x using CNNs and custom clustering
  • Automated solid dispensing and solubility testing robotics in pre-formulation drug discovery
  • Research contributor, High-Throughput Automated Powder Dispensing Platforms (OPRD, ACS)
  • ML R&D consultancy, scaled from solo to 40+ contractors
  • Developed partnerships with Nvidia, AWS, and IBM
  • Production systems across CV, NLP, and time-series for seed-stage to enterprise clients
  • AI speech coaching app — 300K+ users, featured in TechCrunch, Wired, Fast Company
  • Optimized core speech processing pipeline in Kaldi, reducing STT computation by 40x
  • Founder went on to raise $2.3M seed and land Forbes 30 Under 30
  • Boutique AI career consulting born as a Covid-era passion project to give back to the ML community
  • 53 individuals placed into senior+ ML positions through direct 1:1 mentorship
Projects & Research

Research Homelab — DGX Spark, dual RTX 6000, Orin Nano + Pi edge · k3s · OPNsense

  • 45-DOF animatronic of the Speed Racer chimp
  • Maya rig → URDF/USD → Isaac Sim RL → ONNX, C++17 real-time firmware
  • Sim-to-real puppeteering from a Steam Deck
  • Open-source annotation platform — 14 modalities, bounding boxes to LLM RLHF
  • SAM + GroundingDINO assisted labeling, SSO/RBAC, full Docker stack
  • Isometric virtual office for AI coding agents — tmux sessions rendered as characters
  • Three.js + FastAPI/WebSockets, live agent state detection, in-scene terminals
  • Independently replicating PrismML's closed intelligence-density results
  • MoLoRA routing, STE-QAT, speculative decoding, bit-flip continual learning
  • Hierarchy-Aware Hard Negative Mining for Document Retrieval 2026
  • Context-Restricted Prompt Routing for Structured Multimodal Information Extraction 2026
  • Look Sharp: Resolution Scaling for Cartographic VQA SOTA on the FRIEDA benchmark — 44.0% vs 38.2% published best 2026
Upcoming
  • Hierarchy-Aware Hard Negative Mining for Document Retrieval 2026
  • Context-Restricted Prompt Routing for Structured Multimodal Information Extraction 2026
Filed
  • Self-Maintaining Hierarchical Tool Routing for LLM Agent Ecosystems 2026
  • Gated Multimodal Mixture of Experts for Document Classification 2025
  • Auto-Calibration of Projection Systems in Feature Sparse Scenes 2025
Granted
Papers
Older Projects
  • Mobile job-site management platform for construction subcontractors
  • Real-time crew tracking, digital invoicing, and project documentation (getzenhammer.com)
  • Small-batch fashion manufacturer in the Carolina Textile District
  • ML and automation strategy for design-to-production pipeline
  • Technical due diligence and vendor evaluation for ML tooling
  • Designed custom electronics to reduce noise in data, increasing capabilities by 20x
  • Used IoT and signal processing to reduce laboratory costs by 30%
  • Invented and fabricated novel laboratory equipment for solar cell fabrication (Fafarman Group)
  • Supervised 16 engineers designing custom IoT control systems for Drexel's hyperloop pod
  • Pod manufactured and tested at SpaceX HQ in Hawthorne with a team of 110+
  • Synthesized and assayed novel SOD-mimetic compounds (PABA + bacitracin ligands)
  • Presented at the American Chemical Society Conference (2015) — poster
  • Awarded Honorary ACS Membership
  • Distributed botnet platform for automating an MMO at scale
  • Custom scripting engine, anti-detection systems, and multi-account orchestration
  • First real engineering project — everything since started here
Education

BSE Chemical Engineering, Drexel University
Additional studies in chemistry, biology, philosophy, literature, and jazz theory.

Fun
Home

Philadelphia · San Pancho (MX) · NYC · Celebration · Johnson City · Valencia (ES) · Los Angeles

Elsewhere
LinkedIn GitHub