Building the future of manufacturing

Featured: Bed Temperature Fine Tuning Tool

My most widely deployed tool—used by customers, field engineers, and makers like Adam Savage. Diagnostic part with torque spinners, dimpling grids, and birchbark cones for visual/mechanical assessment of thermal uniformity.

Featured: Fuse 1+ 30W Technical Program Management

Despite its "plus" naming, the Fuse 1+ 30W was a complete rebuild of Formlabs' SLS platform—new optical systems, thermal systems, material handling, PCBAs, electronics, and software. I served as Technical Program Manager from EVT through production launch, coordinating 30+ engineers across 130+ ECOs to deliver the program in just 9 months—the fastest full printer launch in Formlabs history.

Program leadership:

• Cross-functional coordination: Owned the engineering program schedule, tracked dependencies across ME, EE, firmware, and software teams. Ran daily syncs to identify blockers and ensure tasks were completed on time
• In-house EVT builds: Coordinated with manufacturing, global sourcing, and the 30+ engineers owning the 130+ ECOs required for this "incremental" update that was actually a ground-up redesign
• Go/no-go decisions: Partnered with the Operational Program Manager to host go/no-go meetings. I owned engineering readiness while translating executive and marketing timelines into achievable engineering milestones

Factory & production:

• Factory test software: Managed the factory software team responsible for the 150+ tests each printer must pass. Owned test design, implementation, and validation—ensuring every unit met spec before shipping
• Production deployment: Worked closely with the factory through daily calls, serving as the point person for test validation and build quality. Led the rollout for months into the production phase
• Release authority: Served as the central decision-maker for production release—determining when printers were ready to ship to customers

Impact:

• The Fuse 1/1+ family now represents over 55% of SLS printers sold worldwide
• The complete redesign moved from prototype through production release in nine months
• Engineering ownership spanned process architecture, requirements, verification, factory software, machine acceptance, and shipment release

Technical Research & Tools

Additional research enabling data-driven decisions for next-generation SLS development:

How the Process Works

Conductive ink is printed into each powder layer, exactly where the part should be. RF energy then passes through the entire build at once, but only the ink-doped powder absorbs it and fuses. Nothing scans and nothing traverses, so a whole cross-section, or a whole part, is heated in parallel. These renders were made with HEATR for a Nazdar press release and are illustrative rather than a specific solved run.

Selectivity, Measured

The renders show the intent. This is the machine doing it: a doped test part under RF, imaged with a FLIR A70 after about two minutes of heating. To the eye it is a dark patch in a bed of white powder. To the camera the doped region is the only thing above room temperature.

Ink Delivery System GUI

I developed a cross-platform desktop application for controlling the APS Engineering NANO 700 ink delivery system used on the RFAM platform. Built with Electron and the Web Serial API, the app provides serial device control, real-time system monitoring, data trending via Chart.js, configurable settings, and a filterable event log—all with full dark and light theme support. It builds for macOS, Windows, and Linux.

Conference Presentation

Solid Freeform Fabrication Symposium 2026 • Austin, TX

Geometry-Dependent Sintering Nonuniformity in RFAM. A talk on HEATR—a coupled electromagnetic–thermal–sintering solver for radio-frequency additive manufacturing—and the simulation-guided compensation framework it enables: graded dopant, exposure scheduling, and a design-for-RFAM capability envelope.

Conference Presentation

Solid Freeform Fabrication Symposium 2025 • Austin, TX

PhD Proposal

Proposal presentation (September 2025)

Research Tasks

• Task 1 – Platform Design & Diagnostics: Custom binder jet-style machine with precision dopant patterning and low-cost scanner-based diagnostic pipeline for jetting fidelity and dopant distribution
• Task 2 – Computational Solutions: Multiphysics FEA simulation of RF heating, functional grading of dopant concentration, geometric pre-warping (ILT-inspired), and turntable-based field averaging for thermal uniformity
• Task 3 – New Materials & Modalities: RF-curable silicones in viscous baths, RF-induced foam expansion, and selective RF actuation for programmable deformation

Key Advantages

• Parallel energy delivery: RF exposure heats doped regions without a rastering laser
• Material-selective heating: Patterned conductive dopant defines where RF energy is absorbed
• Computationally designed process: Coupled physics converts geometry into graded material and exposure instructions
• New machine architecture: The process trades precision scanning optics for RF matching, field management, and dopant deposition

What I Built

• Custom binder jet-style platform with high-precision ball screw actuators (sub-100 μm repeatability)
• Xaar Aquinox 2002 printheads capable of jetting high-viscosity inks (50+ cP) with up to 50 wt% particle loading
• 600W, 13.56 MHz RF generator (AG0613) with AIT-600 automatic impedance matching network
• Custom Faraday cage with parallel copper electrodes and motorized turntable for isotropic field exposure
• RF-transparent build chambers (PLA/nylon) with compliant printed spring sealing systems
• Custom high-loading graphite inks (up to 25 wt%) developed with Nazdar

October 2026: A Hand-Built Matching Network, and the First Sintering Run

The automatic tuner could not hold a match into a load whose impedance moves as the powder melts, so I built the matching network by hand and instrumented it. That work turned up a failure that had been corrupting every thermal measurement on the machine, and it ended with the first part this platform has fused.

Chasing an RF Leak to Its Source

The symptom was that the FLIR camera dropped out under power and the thermal images carried RF noise. The cause was mechanical. The coax entered the Faraday cage through a plastic-lined wall slot and its connector hung on zip ties, making no metal contact with the cage, so the shield was never bonded to the wall. The network's secondary sits at kilovolts, and the stray capacitance from that secondary to the cage carries a current that has to get back to the generator. With no bond at the wall it returned over the outside of the coax and the ground wire, which put RF on the cage, its frame, and the camera mount. The fix was a braid strap from the coax flange to the cage wall.

Instrumenting the Network Itself

With the camera trustworthy again, the same thermal imaging could be turned on the network. At the end of a 300 W hold the hottest spot was the circuit board beside the first high-voltage post at 90 °C. Scaling that rise with power puts the same spot at 118–135 °C at 500 W, against an FR4 softening point near 130 °C, so the board becomes the binding constraint before the generator does. That is a projection from the 300 W measurement rather than a measurement at 500 W, and it is why the next run targets a constant 250 W rather than pushing the top end. RF interference in the image stayed between 0.9 and 1.6 mK at every step of the sweep.

The First Fused Part

A 25.4 mm doped cube was ramped to 300 W. It collapsed at 553.7 s and smoked from its center about 75 seconds later. What came out of the bed was a fused body of 59.8 × 40.5 × 16.5 mm weighing 20.6 g, of which roughly 12 g was undoped nylon that had never been printed with dopant at all. Melting had run 9 mm across the field and 17 to 21 mm along it.

That extent is the interesting part. A quasi-steady melt-front estimate says conduction alone would need the cube face held somewhere between 400 and 700 °C to melt 19 mm along the long axis, and nylon degrades above about 350 °C. So conduction does not explain the part. Two possibilities remain open: RF heating of the hot undoped nylon once it is above its own transition, or doped powder that spilled into the surrounding bed. Distinguishing them is the next experiment, not a settled result.

Advisors

Dr. Carolyn C. Seepersad (Woodruff School Chair & Professor) & Dr. Christopher J. Saldaña (Ring Family Professor)

See it in motion

A short field guide to HEATR — the electrothermal pipeline (field solve → RF power → melt → densification), then launching a run and reading the results.

From Geometry to Printer-Ready Compensation

The second-generation solver minimizes shape error directly, penalizing both unfused material inside the target and melt that escapes into the surrounding powder. One backward adjoint sweep returns the gradient for every dopant cell at roughly the cost of one forward solve. The pipeline jointly selects the dopant field and exposure stop, then filters and quantizes the result to the machine's 4-bit raster convention before re-running it through the production forward model.

Across the standardized 18-shape library, the printable single-pass map beat uniform dopant on all 18 shapes and the best stored historical map on 13. Seven shapes reached the absolute solved class of melt-region IoU ≥ 0.95. Four-bit quantization changed IoU by no more than 0.0035 across the census, showing that the optimized structure survives the transition from a continuous design variable to a printable instruction.

The Geometry-Coupling Problem

In a parallel-plate RF field, energy coupling is not uniform. Corners and edges perpendicular to the applied field concentrate the E-field dramatically, while surfaces parallel to the field couple weakly. This means every different geometry produces a different heating pattern, and for non-symmetric shapes the results can be severe: one region of a part can be fully sintered while another barely reaches melt temperature. This is not a bug you can engineer away at the system level. It is an inherent feature of EQS heating. The only path forward is to understand the coupling for each geometry and develop compensation strategies that work around it. That requires a simulation tool fast enough to explore the space systematically.

Why Build a Custom Solver?

COMSOL can assemble the same coupled physics, and it remains an independent reference for HEATR. The advantage is not an exaggerated single-solve speed claim: a comparable transient case takes roughly 30 seconds in HEATR versus 60–70 seconds in COMSOL. The decisive difference is throughput and control. HEATR carries no per-run meshing, GUI, or license-seat overhead, so hundreds of unattended geometry, orientation, grading, and adjoint evaluations are scriptable. Owning the forward model also makes it possible to embed custom densification endpoints, quantization gates, inverse-design objectives, and exact adjoints directly in the solve.

Current Design System

• Forward physics: Coupled EQS, RF power deposition, transient thermal transport, phase change, and densification in 2.5D and 3D.
• Adjoint material design: Shape-fidelity gradients for every dopant cell at approximately one additional forward-solve cost.
• Printability gates: Filtering, smoothing, 4-bit quantization, grid transfer, energy-residual checks, and production-forward re-evaluation.
• Additional actuators: Exposure stop, RF drive, orientation, asymmetric dwell, field averaging, and geometry pre-distortion.
• Method selection: Geometry-specific routing identifies where grading works and where field shadowing or power limits require another lever.

Watch the Inverse Design Converge

The animation below contrasts a one-shot proportional inverse mask with the adjoint solve. Instead of inferring a correction from a proxy field, the adjoint repeatedly evaluates the printable dopant map through the coupled RF, thermal, and densification physics, driving the final melted geometry toward the target.

When Material Grading Is Not Enough

Some re-entrant and multi-lobed geometries remain limited by field shadowing or available power. HEATR can then search process actuators as well as material maps. These SFF 2026 animations show how rotation changes the heating kernel, how an indexed two-stage schedule melts an L-shape limb by limb, and how asymmetric dwell distributes energy across a cross. The results are simulation-established; the current electrode-medium contact architecture prevents physical turntable implementation without a hardware change.

Verification, Cross-Checks & Open Validation

Every run reports field maps, thermal and densification histories, energy-balance residuals, and numerical diagnostics. The forward physics is cross-checked against COMSOL on reference cases, and a one-parameter calibration against a recovered independent IR transient reproduces the measured time-to-melt. The adjoint is verified against finite differences before optimization and satisfies the forward/adjoint transpose test to machine precision.

The remaining boundary is explicit: dedicated full-field validation on the current RFAM machine has not yet been completed, and the model still uses published 30 wt% graphite-in-nylon electrical data as a stand-in for the deployed 25 wt% carbon-black ink. Absolute heating magnitudes therefore remain model-relative pending impedance spectroscopy and closed-loop IR validation. The compensation results are simulation-established, not yet experimentally demonstrated.

Simulation Platform & GUI

HEATR includes a browser-based GUI for operational control and results exploration. The Operation page provides run configuration (geometry, mode, exposure, turntable parameters), a live job queue with real-time progress tracking, and inline simulation previews. The Results browser offers filterable run history with hero previews, expandable per-run metrics with contextual explanations, and quick access to all output figures. The platform also includes a Files tab for direct output browsing and a Theory tab with integrated physics reference documentation.

What the Inverse Design Buys You

The clearest way to see what grading does is to print the same hexagon twice. Fill it with ink uniformly and the RF field over-couples at the top and bottom and starves the flanks, so the part melts into an ellipse. Let the solver grade the ink instead, within the eight grey levels the printhead actually has, and the melted region comes out as the hexagon that was asked for. Shape match goes from 76 to 97 percent. Both are simulation.

October 2026: Running the Model Against a Real Part

A thermal camera only ever sees the top surface. In a volumetric process that is the one place where nothing interesting happens, because the whole point is that the part heats from within. The first instrumented sintering run was a chance to ask what the model says is underneath while the camera watches the top.

Through the run the camera's 24 mm box mean climbed to roughly 100 °C and never approached the 171 °C melt onset. Over the same window the model's core-center band runs to about 330 °C, well past melt. The part then collapsed at 553.7 s and smoked from its center at about 628 s, which is the behaviour of an interior that is far hotter than its surface.

The honest reading matters here. The cube's heating law was fitted to the camera, so the agreement on the top face is partly by construction and is not independent evidence. The core trace is an extrapolation from that fit, and the collapse and the smoke are physical corroboration of it, not a validation of it. The model is not valid after the collapse at 553.7 s, and this run carried no densification and no RF loss in the undoped powder. What the run does establish is that the surface reading alone would have been badly misleading, which is the case for having the forward model at all.

Technical Details

• Physics: Complex-coefficient electro-quasi-static field solve, generator-power-scaled RF deposition, transient thermal transport, DSC-calibrated apparent heat capacity, and viscous-capillary densification
• Models: Production 2.5D solver plus a volumetric 3D port for full build-height studies
• Inverse design: Direct/adjoint material optimization, exposure-stop selection, orientation, field averaging, and geometry pre-distortion
• Numerical gates: Finite-difference gradient checks, grid refinement and transfer checks, filtering, 4-bit quantization, energy residuals, and production-forward re-evaluation
• Stack: Python, NumPy, SciPy (sparse), Matplotlib, browser-based GUI with live job management

Two Diagnostic Capabilities

This tool serves two complementary functions for RFAM process validation:

• Dimensional Accuracy: Evaluates jetting fidelity and compensates for system-level mechanical errors (scale, yaw, backlash)
• Ink Concentration Analysis: Characterizes ink deposition properties—from interactive ROI-based measurements through Bayesian classification and Gaussian Process regression—to verify dopant distribution for RF heating uniformity

Dimensional Accuracy Analysis

Calibration patterns are printed on low-wicking substrates to preserve precise droplet locations, then scanned and analyzed to extract geometric error metrics.

Calibration Patterns

• Dot arrays: Centroid-based pitch error, diameter, circularity, eccentricity, and grid rotation
• Checkerboards: Sub-pixel corner detection, pitch estimation, and yaw compensation
• Concentric rings: Radial symmetry, linewidth/spacing extraction, edge fidelity stress-test
• Pitch rulers: Minimum resolvable bar width as a function of sampling resolution

Ink Concentration Analysis

For dopant concentration analysis, chromatography or blotting paper substrates are used due to their ability to absorb and spread droplets into consistent, analyzable halos. Users interactively select regions of interest using polygon, circle, and ruler tools, then the pipeline segments ink from substrate via Otsu thresholding and extracts a rich metric suite: intensity statistics (mean, std, IQR, skewness, kurtosis, entropy), shape descriptors (circularity, convexity, inertia ratio), and halo eccentricity via morphological erosion and ellipse fitting.

Multiple analysis modes are available for identifying ink type and predicting carbon loading. Deterministic classifiers (weighted score, k-NN, logistic regression) handle material identification and concentration binning, while a Bayesian Gaussian Mixture Model classifier provides posterior probabilities with entropy-based uncertainty quantification. For continuous concentration estimation, Gaussian Process regression with a Matérn 2.5 kernel produces calibrated predictions with confidence intervals, alongside isotonic and k-NN regression baselines. Together these tools verify that the correct dopant concentration is being deposited across the powder bed—critical for achieving uniform RF heating during sintering.

Key Contributions

• Digital "gold-standard" validation isolates algorithmic behavior from physical printing variability
• Normalized circularity metric (C_norm) referenced to intrinsic discretization ceiling (C_∞ = 0.899)
• Structured dimensional health report separating kinematic compensation from morphology diagnostics
• GUI and CLI with standardized outputs (results.csv, raw.csv, debug overlays) for cross-session tracking
• Open-source pip-installable Python package (pip install bjam-toolbox) for community use and reproducibility

Diagnostic Workflow

• Pattern definition → Rasterization/scanning → Feature extraction → Error processing → Compensation
• Separates compensation-driving metrics (scale, yaw) from process health indicators (circularity, eccentricity)
• Produces actionable recommendations for motion calibration and process tuning

Publication

Manuscript in preparation

Journal Publications

Conference Papers

Planned 2027 Journal Portfolio

• RFAM machine architecture and polymer powder-bed process development
• HEATR coupled electrothermal and densification solver, verification, and experimental cross-validation
• Adjoint inverse design of printer-ready functionally graded dopant maps
• Geometry compensation and pre-distortion through coupled process physics
• Low-cost scanner metrology for jetting fidelity and dopant distribution

Patents & Disclosures

• Patent contributions/invention disclosures at Formlabs (SLS systems and processes)

Full Demo Video

Manufacturing & Build Process

Quick Highlight

Media Coverage

Concept

The installation simulates an interdimensional portal that has "ripped" through a wall, revealing glimpses of alternate futures. Guests can use two physical dials to control the experience: one selects the alternate world, and one selects the campus location they're viewing. Five distinct worlds were created, each with three campus locations (15 total scenes):

• Digital World (2092) - A cyberpunk future with neon-lit holographic interfaces
• Green World (2075) - Nature has reclaimed the campus with overgrown vegetation
• Outer Space (2524) - Georgia Tech as an interstellar space station
• AI Takeover (2148) - Robots and AI have transformed the campus
• Underwater (2262) - The campus submerged in an underwater city

Installation Videos

Process & Exhibition

Installation Overview

Control Panel

Portal Frame

World Selection

Guest Interaction

Alternate Reality

Physical Installation

Exhibition Setup

AI-Generated Content

Final Exhibition

Technical Implementation

• Hardware: Arduino-controlled rotary encoders and physical dial controls
• Software: TouchDesigner for real-time video playback and UI feedback
• Content: AI-generated video content using Runway, with post-production in After Effects and Premiere Pro
• Physical Design: Drywall wall construction with sculpted foam "broken wall" effect and rear projection surface

The Problem

As populations age globally, many seniors face challenges staying physically and socially active. Research shows that maintaining regular activity is crucial for both physical health and mental wellbeing, yet many older adults struggle to find and plan appropriate activities that match their preferences, abilities, and social needs.

Interactive Prototype

Click through the app to experience the onboarding flow and main features.

Key Features

• Personalized Onboarding: Questions about activity preferences, social styles, travel willingness, and schedule to tailor recommendations
• Weekly Activity Planning: Calendar-based interface for scheduling activities throughout the week
• Social Connection: Features to coordinate activities with friends and meet new people with similar interests
• Organizer Portal: Separate flow for activity providers to list and manage events
• Accessibility Focus: Large text, simple navigation, and clear visual hierarchy designed for older adults

Design Process

• Conducted interviews with elderly community members and healthcare workers
• Created journey maps and service blueprints to identify pain points
• Developed personas based on research findings
• Iterative prototyping with user feedback
• Built working prototype using Figma and React

Instruments

Recognition

• NIU Music Scholar
• Louis Armstrong Jazz Award
• Purdue Jazz Festival Winner
• Lead Alto - NIU Jazz Orchestra
• Lead Alto - NIU Jazz Ensemble

Mentors

Geof Bradfield, Reggie Thomas, Rich Moore