Topic · The broader APT.ai portfolio
What elseAPT.aiis doing.
GUS.ai is one initiative inside an AI-native engineering firm. The work that follows sits beside it. Some is shipping today. Some is in pilot. Some is research on the bench. Stage is labeled on every project so a multi-year engagement stays honest.
Chapter 12
The
portfolio.
What else APT.ai is building. The work that sits beside GUS.ai.
Already running · 1 of 2
Agents and platforms. Live today.
Running right now on cloud droplets and local workstations. The engineering platform that sits beside GUS.ai.
OpenClaw
Running
Autonomous AI agent on cloud infrastructure. Communicates via Telegram, sends email through a sandboxed channel, queries GUS.ai for live telemetry. 13 skills including GUS tools, intelligence feed, research briefs, build coordination.
APT Knowledge Base
Running
Standalone application that ingests engineering textbooks, PDFs, research papers, and standards into a searchable vector store. Exposed as an MCP tool so GUS.ai, OpenClaw, and any future agent can query the full library.
AI Council
Designed
Six specialist AI advisors covering engineering, business, technology, competitive intelligence, finance, and customer voice. Meet twice daily on rotation. Synthesize a memo to leadership. Cents per session.
Intelligence Feed
Running
Autonomous research agent. Scores industry newsletters, generates deep-dive briefs on competitors, technology, and market trends. Pushes findings to an internal research store. Continuously up to date.
GUS Vision
Built, testing
Browser-based real-time computer vision for cheese-vat monitoring. RF-DETR bounding boxes plus Claude scene narration, running entirely in-browser via WebGPU. No backend, no API keys, no model training. Operationalizes published curd-syneresis color-shift research.
Sales Intelligence Platform
Designed
Automated competitive intelligence and lead scoring for the sales team. Crawls 40-plus sources across 8 categories. Weekly Monday 7am digest. Three intake channels: email, SMS, voice. Seven-factor weighted scoring with human boost.
Already running · 2 of 2
Twin, capture, and on-device AI.
The physics, the plant-walk, and the local-only stack. Same engineering team.
ACV Digital Twin
In progress
Multiphysics simulation of the Advanced Cheese Vat. Blade CFD, heat transfer through dimple jackets, rennet coagulation kinetics. SolidWorks geometry plus NVIDIA PhysicsNeMo surrogates. Foundation for RL training and virtual commissioning. Detail in the next panel.
GUS Vantage (PUPS engine)
Phase 1 build
Walk a customer plant with an iPad Pro, get a navigable 3D Gaussian Splat in the browser within 30 minutes of capture. Targets the consumer-hardware path that replaces 50K dollar laser scanners for plant scoping, retrofit, and tie-in design.
GUS Spark
Scaffolding
Sketch-to-photoreal adaptive rendering. Engineer draws a vessel or piping isometric on iPad with Apple Pencil, sees a photoreal render in under 250 ms. Local SDXL-Turbo plus ControlNet on a Blackwell GPU, no cloud dependency.
APT FormScan
Running
Vision pipeline that converts handwritten customer QC log books (vat milk samples, line-check forms, batch logs) into reviewed Excel workbooks. Severity-colored outlier flagging, process-observation notes, one tab per form page. Each customer's form is a self-contained template.
NemoClaw
Built
Local-first AI personal assistant on a Blackwell GPU workstation. Nemotron inference fully on-device; privacy router keeps sensitive APT engineering data off the cloud. APT's internal sandbox for sensitive R&D, not a customer deployment option.
The agent fleet
Five specialist agents. Each with a discipline.
GUS.ai is the conversational layer customers see. Behind it, named agents target specific APT engineering disciplines. Each can be spun up for an individual workflow in hours, not months.
SPECIA
Critical
Specification Validation and Authoring Agent. Two modes: validation (check BOM against manufacturer catalogs) and authoring ("I need an ECTFE system for 100 gpm caustic at 60 F" returns a complete spec with part numbers). Replaces hours of catalog lookup per project.
CRUX
Running
ADX AI/ML Architect. System design, configuration, troubleshooting for ADX. Knows the full microservices stack: Ignition Edge, EventGateway, HiveMQ, Chronicler, Museum, PLCTagWriter. Translates requirements to ISA-95 schemas.
BESS-E
Running
Batch Engineering and Sanitary Systems Expert. Generates and critiques P&IDs. Selects pumps, valves, heat exchangers. Flags contamination risks. Ensures 3-A, FDA, USDA compliance.
LATTICE
Running
Whey Processing and CFD Expert. Analyzes CFD results: velocity fields, vorticity, turbulence. Recommends impeller designs and baffle strategies. Spans membrane separation, evaporator optimization, and crystallization where applicable.
CHIMERA
Running
Chemical Hazard and Interaction Mapping Engine. Evaluates chemical compatibility, reactivity risks, material safety across the dairy and food-contact stack.
GUS Agile · the deep dive
AI-assisted agitator design. Measured, calculated, cited.
The first GUS Agile engagement shipped against a regional dairy customer's 75-percent fat cream storage tanks. Every number traceable to a literature correlation or a measured lab value.
8
Tanks designed
2,500 gal
Tank capacity
1.86 m/s
Tip speed
2.1×
Motor margin
82 min
Heat-up time
99.9%
Product recovery
APT cream-tank engagement reference · methodology generalizes to blending, reaction, CIP duty
GUS Agile · the methodology
Rheology to recovery. End to end.
Four stages, in order. The platform measures, calculates, sizes, and reports — with every step cited against literature or lab.
01
Measured rheology
Independent lab characterization (power-law or Newtonian) of the actual product, at the actual operating temperatures. The platform does not assume; it measures.
02
Power and shear calc
Metzner-Otto effective shear rate, Holland-Chapman power correlations, scraper drag estimation, tip-speed and Reynolds validation against literature bounds.
03
Heat transfer prediction
Penetration-theory inside film coefficient for scraped jackets, batch heat-up time simulation, thermal margin sized against operating envelope.
04
Recovery and design rationale
Scraper geometry, motor margin, baffle strategy, and product-recovery prediction collapsed into a branded engineering report. Reproducible across duty classes.
The deep dive
A vat that thinks back. Live physics, live telemetry.
The most visually concrete project on the bench. A reference Advanced Cheese Vat, real telemetry, three physics models running together at interactive frame rates. We show it because it answers “where is the AI” more directly than any chart can.
Blade CFD
Navier-Stokes around the agitator
Velocity streamlines, shear stress, vortex structures. Updated live as agitator RPM changes on the floor. The operator sees the same flow field the rennet does.
Heat transfer
Steam jacket to bulk fluid
Conjugate thermal model from the steam jacket through the vat wall into the bulk. Radial temperature gradients driven by live cook water and vat temperature, not a static profile.
Rennet coagulation
Reaction-diffusion in real time
Enzyme-driven casein coagulation modeled as a reaction-diffusion field. Concentration front propagation responds to mixing intensity and temperature, batch by batch.
Tech stack: NVIDIA PhysicsNeMo for surrogate training (PINN, neural operators). OpenFOAM for GPU-accelerated CFD ground truth. ONNX Runtime for live inference. WebGPU and Three.js on the front end. Geometry from SolidWorks STEP export. Telemetry from the same TimescaleDB GUS.ai uses.
ACV digital twin · reference deployment · pilot
What a partnership looks like
GUS.ai is one engagement. The relationship grows.
A typical engagement starts with one tool. Follow-on work might include data consolidation, vision-based form digitization, lab adapter work, or a domain-specific investigation. Each scope is independent; the team is the same.
AI-native engineering firm
Build, ship, operate. Not a chatbot vendor. The team that wrote the GUS.ai backend also writes the digital-twin physics, the form-digitization vision pipeline, and the cross-source RCA agents. One discipline, applied across products.
Multi-product, single relationship
GUS.ai is one engagement. The lab adapter, the form digitization, the digital twin, the piping work all sit beside it as separate scopes. Same team, same primary contact, multiple parallel workstreams.
Workflow-restructure over kiosk-install
Where AI capability lands a meaningful productivity lift, it's because a workflow restructured around the agent — not because a tool got installed in the corner. The portfolio above is the work that follows when one ritual restructures.
End of topic · Other projects