Designed and developed five electromechanical ocular surgical instruments spanning minimally invasive retinal detachment repair, subretinal drainage, suprachoroidal space (SCS) drug delivery, glaucoma treatment, and subretinal delivery — bringing each from initial concept through prototyping to clinical readiness.

Invented core technology resulting in 5+ patent applications and which enabled the company to raise over $10 million to bring the retinal detachment device through clinical trials and to market. Led end-to-end development of the lead device from conception through preclinical animal studies to first-in-human clinical use within approximately 1–2 years.

Managed preclinical and first-in-human studies, including protocol design, operating room support during 10+ procedures, data analysis, and direct collaboration with surgeons to iterate on instrument design. Contributing to regulatory submissions including IDE and IRB filings. Managed 4+ contract manufacturers across different specializations to advance commercialization.

Comparison of current ocular drug delivery routes (invasive, untargeted) versus the novel minimally invasive, targeted approach developed at Dragonfleye.

Technical drawings showing internal mechanism design — spiral cam actuation, needle holding member, pivot points, and tip needle assembly across multiple instrument views.

From CAD to clinic — rendered models (A–C) alongside functional prototypes (D–F) including the transparent development prototype and control unit with foot pedal activation.

Endovascular surgery uses steerable catheters inserted through the femoral artery to treat conditions like aortic aneurysms and stenoses. While minimally invasive, surgeons face challenges with imprecise manual catheter control and poor 3D visualization under X-ray fluoroscopy. This project explored two approaches to robotic endovascular catheter control, both guided by 3D intracardiac ultrasound imaging as an alternative to radiation-based guidance.

Designed a fully steerable robotic catheter system with 4 degrees of freedom — pitch, yaw, roll, and translation — for semi-autonomous aortic navigation. The system uses a tendon-driven bending mechanism with reciprocal spools to control catheter tip orientation, integrated with a Philips VeriSight Pro 3D intracardiac echo catheter for real-time vascular imaging. The goal is autonomous catheter navigation using SLAM algorithms under surgeon supervision.

The four required degrees of freedom for the ultrasound catheter system: pitch and yaw for tip steering, roll for 360° ultrasound imaging, and translation for advancement through the vasculature.

System layout showing bending, roll, and advancing modules on a height-adjustable platform, with the steerable outer catheter housing the ultrasound catheter and an introducer for vessel access.

The bending module uses four tendons routed through a straightening module with adjustable pulleys, then wrapped around reciprocal spools. Each spool couples opposite tendons so that one motor controls each bending axis — tensioning one side while releasing the other to bend the compliant catheter tip without buckling.

CAD design of the bending module: reciprocal spool and catheter holder assembly with support bearings (top), and full bending module with straightening pulleys, motors, and catheter clamp (bottom).

Physical prototype of the bending module with 3D printed catheter holder, laser-cut acrylic motor mount, DC motors with encoders, and the reciprocal spool mechanism for tendon actuation.

Validated the bending module using video tracking with fiducial markers to correlate motor encoder position with catheter tip deflection. The system demonstrated greater than 90° bending range in both planes with a mostly linear motor-to-catheter angle relationship, confirming precise tip position control.

Bending module testing: catheter in straight position (top) and deflected during actuation (bottom). Red fiducial dots enable video tracking of tip angle via circle-fitting in Tracker software.

Motor angle (black) and catheter angle (orange) vs. time across six trials in vertical and horizontal planes. The close correlation demonstrates high sensitivity and a large range of motion exceeding ±90°.

Catheter angle vs. motor displacement during bending resolution experiments in vertical and horizontal planes. The nearly linear relationship confirms reliable position control with some hysteresis near zero degrees.

3D printed phantom aorta model for testing semi-autonomous catheter navigation from the main aorta into renal artery sub-branches using SLAM algorithms.

The first-generation device took a different approach: rather than building a fully steerable catheter, it enables surgeons to seamlessly switch between manual and robotic control of a standard off-the-shelf catheter mid-procedure. A friction drive module with DC motors and spring-loaded idler rollers engages or releases the catheter via a push-button latch — providing robotic precision for critical maneuvers while preserving manual tactile feedback for routine navigation.

V1 system overview: catheter navigating a vascular phantom, friction drive module on a linear stage, and Philips 3D ultrasound machine providing real-time imaging guidance.

Free body diagram of the three-roller friction drive. The drive roller (D1) provides traction while idler (D2) and compression (D3) rollers maintain grip without damaging the catheter.

Friction drive module in open configuration showing drive rollers, idler rollers, DC motors, and push-button release latch for rapid catheter insertion and removal.

Device in closed configuration with catheter engaged. Compression springs maintain consistent roller pressure for reliable friction drive without slippage.

FEA of roller-catheter contact showing Von Mises stress distribution. Results confirm catheter structural integrity under operating grip forces.

Catheter knob rotation vs. roller angular displacement. Linear relationship confirms reliable rotational control for catheter steering.

Designed and built a multifunctional competition robot from a constrained kit of materials and actuators as part of MIT's legendary 2.007 Design and Manufacturing course. The robot competed head-to-head against 150 students in a themed arena challenge requiring autonomous and teleoperated tasks.

The robot featured a multi-linkage arm with a custom gripper mechanism for manipulating game objects, belt-driven drive wheels with DC motors, and integrated electronics for motor control and sensing. All structural components were designed in CAD, machined on the mill and lathe, and assembled under tight time constraints.

Won the International Design Competition Prize — the highest award in the class — recognizing overall excellence in mechanical design, fabrication quality, strategic thinking, and competition performance.

Expandable scissor arm mechanism using masonite bars, rubber bands, and a string-and-spool pulley system driven by servos. The arm extends to reach elevated game objects while remaining compact when retracted.

Early design sketch showing the unfolding arm concept with pulley system, servo-driven rotation, and grabbing tool. The ~30 cm arm was designed to reach up to the zipline and paint cans in the competition arena.

Front view of the competition robot showing the belt-driven drive system, DC motor, and articulated arm mechanism in retracted position.

Overview showing the complete robot assembly — chassis, drive motors, arm mechanism, and wiring harness. All structural components machined from the constrained kit materials.

Extended arm mechanism showing the full reach of the multi-bar linkage. The arm uses a combination of rigid links and cable-driven actuation for compact, powerful deployment.

Side profile of the arm at full extension, showing the linkage geometry, pivot points, and the relationship between the drive mechanism at the base and the end effector at the top.

Close-up of the central linkage mechanism showing the machined aluminum links, shoulder bolts at pivot points, and the cable routing for actuation.

Detail of the motor mount and drive assembly showing the DC motor, coupling mechanism, spool for cable actuation, and precision-machined mounting hardware.

Custom end effector / scoop designed for manipulating game objects. Sheet metal formed and fastened to create a lightweight, functional gripper geometry.

Investigated the relationship between cleat pattern geometry and traction during soccer-specific cutting maneuvers. Designed seven cleat patterns — including novel concentric arrangements around the center of pressure — and built a custom test system to measure both rotational torque (associated with ACL injury risk) and linear traction (required for performance). Defined a new parameter, β, as the ratio of linear to rotational traction to identify optimal designs that maximize performance while minimizing injury risk.

(a) Custom test apparatus on artificial turf with weighted foot and tripod legs. (b) Diagram of rotational and linear traction measurement methodology. (c) Seven 3D-printed cleat patterns: round, bladed, sharp, double CoP, concentric midfoot bladed, concentric midfoot triangular, and single CoP/TNT.

Custom Cleat Design

Hypothesized that arranging studs in concentric rings around the center of pressure — rather than in traditional distributed patterns — would reduce rotational resistance while maintaining linear grip. Designed all patterns in CAD and 3D-printed prototypes for testing on artificial turf.

CAD model of the concentric cleat pattern showing studs arranged in circular arcs around the center of pressure, with dotted lines indicating the rotation paths that minimize resistance to turning.

Test Apparatus

Designed and built a precision test rig using 80/20 aluminum framing, pneumatic cylinders for controlled normal loading, and instrumented fixtures for measuring both rotational torque and linear traction force on real artificial turf.

Complete CAD assembly of the traction test system with 25 labeled components, including aluminum frame, pneumatic loading cylinder, linear rails, load cell, and interchangeable cleat foot plate.

Results

Tested all seven patterns and found that the single CoP/TNT design achieved the lowest rotational torque (0.72 Nm) while maintaining high linear traction (120 N) — yielding the best β ratio.

Rotational torque (left) and linear traction force (right) for all seven cleat patterns. The Single CoP/TNT design achieved the lowest torque while maintaining competitive linear traction.

Torque vs. traction scatter plot with β = 1 line (dashed). The Single CoP/TNT design (dark red) sits in the optimal lower-right region — high linear traction with minimal rotational resistance.

Human Trial Validation

Conducted human trials comparing three cleat configurations during soccer-specific cutting maneuvers on artificial turf. Results confirmed that the concentric designs reduced rotational traction while maintaining equivalent linear performance.

Human trial results: rotational traction (left) and linear traction (right) for three configurations during cutting maneuvers. Statistically significant reduction in rotational torque (* p < 0.05) with equivalent linear performance.

High-speed photography provides visual storytellers the tools to produce dramatic imagery and gain scientific insight into animal phenomena that occur too fast for the naked eye. However, bringing high-speed cameras into a marine environment presents unique challenges: large power supply requirements, massive data storage, and critical prevention of instrument overheating. This project developed a first-prototype submersible housing for a high-speed camera system with a focus on thermal regulation.

The housing fits an IDT Vision OS 10-4K camera filming at 1,000 FPS, is watertight to 10 meters depth, and maintains temperatures below 55°C while the camera outputs 48 watts of heat in tropical water. Focus adjustments and triggering are possible when sealed, and the camera can be removed from the housing for servicing.

Designed a continuous aluminum thermal path from the camera to the surrounding water while minimizing thermal resistance. Plates on top and bottom of the camera connect directly to the rear plate. Heat pipes using phase-change of distilled water are inlaid along the bottom and back plates. The 1.5" thick rear plate features machined fins to maximize water contact area.

Full housing assembly CAD render showing transparent polycarbonate tube, dome port with wide-angle lens, internal camera mounting system, and finned aluminum rear plate for heat dissipation.

Thermal simulation showing temperature distribution from camera electronics (max 55.87°C at mounting plates) through the heat management system to the water-cooled rear housing (min 29.19°C at the finned end cap).

Fabricating the rear plate required starting with a 9" diameter cylinder of aluminum, turning it down on a CNC lathe, then programming all other features in CAM to be cut on a mill. To machine the cooling fins, the part was cut halfway through on one side, then flipped and cut from the other side.

CNC lathe turning the 9" aluminum cylinder for the rear plate — the most complex component, requiring operations on both lathe and 3-axis mill to achieve the finned heat-sink geometry.

Finished rear end cap showing CNC-machined cooling fin slots for maximizing surface area in contact with surrounding water, and the central camera mounting features.

Successful water tank testing of the complete housing assembly, validating watertightness and heat transfer performance with load resistors simulating camera thermal output.

Collaborated with a vascular surgeon at Beth Israel Deaconess Medical Center to design and develop a pneumatic diagnostic device for Popliteal Artery Entrapment Syndrome (PAES) — a condition where the popliteal artery behind the knee is compressed during physical activity, restricting blood flow to the lower leg.

Designed a ball-screw-driven foot plate actuation system with pneumatic force application, Arduino-based control logic, and a custom GUI for real-time force and displacement monitoring. The device applies controlled dorsiflexion forces to the foot while measuring vascular response, enabling non-invasive diagnosis of PAES without costly imaging.

Published the work in the ASME Journal of Engineering and Science in Medical Diagnostics and Therapy (JESMDT). Presented a poster at the ASME Design of Medical Devices conference. The device was validated in clinical testing at Beth Israel with real patients.

Early concept sketches exploring linear and rotational actuation approaches for applying controlled dorsiflexion forces to the foot plate while the patient lies on a hospital stretcher.

CAD assembly of the final device design: (a) pneumatic cylinder, (b) linear actuation mechanism with foot plate rail, and (c) complete system mounted beneath a hospital stretcher.

Four pneumatic operation modes analyzed for the device: constant force (with relief valve), single-sided air spring, double-sided centered spring, and double-sided off-center spring — each with corresponding force equations.

Force vs. displacement curves comparing all four pneumatic modes. The double-sided off-center configuration was selected for its progressive resistance profile that best mimics physiological calf muscle loading.

Custom pneumatic circuit board with pressure regulators, ball valves, and quick-connect fittings. Connects to hospital compressed air supply and routes controlled pressure to the double-acting pneumatic cylinder.

Complete PAES diagnostic system deployed in the orthopedics department at Beth Israel Deaconess Medical Center — pneumatic circuit on the stretcher, foot plate mechanism mounted below, ready for patient testing.

Clinical testing with a patient's foot engaged against the foot plate. The pneumatic cylinder applies controlled dorsiflexion forces while sensors capture displacement and load data in real time.

Close-up of the CNC-machined linear rail mechanism showing the aluminum foot plate carriage, linear bearings, and precision ground shafts that guide the pneumatic actuation with minimal friction.

As part of NASA's Artemis program, compact deployable composite boom towers are being developed to support lunar surface exploration. These lightweight towers can elevate payloads like solar arrays, navigation beacons, and imaging systems up to 16.5m above a lander deck, providing critical line-of-sight to nearby robotic assets exploring permanently shadowed regions near the lunar poles.

However, deployable composite booms exhibit natural axial curvature that causes lateral tip deflection, limiting tower height and payload capacity. This project designed, built, and tested a deployable guy wire stability system to actively correct these deflections using differential tension control. Static tests at heights up to 8.5m demonstrated that the guy wire system can reduce boom tip deflections by up to 63%.

The SELTI tower uses a collapsible tubular mast (CTM) boom from NASA's Deployable Composite Booms program — a 13m carbon fiber structure with two omega-shaped thin shells that rolls flat for compact stowage and deploys into a stiff tubular cross-section. The tower self-levels in lunar gravity and is envisioned to elevate CubeSat-scale payloads from a lander deck.

Left: The MELLTT proof-of-concept tower with motorized deployer, boom bracing, and leveler plate. Right: The CTM boom in deployed state showing omega-shaped shells, and in stowed state showing thin carbon fiber layers that collapse and roll flat.

Designed a modular three-arm guy wire structure with deployable spreader arms, each equipped with a brushless motor for precision tensioning, load cells for real-time force measurement, and a ratchet-and-pawl mechanism for passive tension locking. CNC-machined aluminum components and 33 press-fit ball bearing joints ensure structural stiffness and rigid-body behavior.

Top: Guy wire arm components — tensioner motor, ratchet release actuator, deployment actuator, truss, pulleys, and load cell. Bottom: Three arms mounted on the deployer-leveler interface plate, viewed from above.

Complete guy wire system arrangement: three arms spaced 120° apart mounted to the deployer base, with the boom extending vertically. Inset (a) shows the load cell force measurement path; inset (b) shows the ratchet and pawl detail for passive tensioning.

Conducted static tests in the MIT Stata Center stairwell at three deployed heights (4.2m, 6.2m, and 8.5m) using an OptiTrack photogrammetry system to measure boom tip position under various guy wire tension configurations. Validated the photogrammetry system to sub-millimeter accuracy against conventional ruler measurements.

Boom deployer from above showing the three guy wire arms (labeled 1, 2, 3) at 120° spacing, the green deployment rollers, and the X-Y coordinate system aligned with the boom cross-section seams.

Photogrammetry results at three heights. Left column: single-arm tension on arm #3 progressively reduces Y-axis deflection. Right column: differential tension across all three arms brings tip position closer to the base center, demonstrating effective 2D control capability.

Results confirmed that tension-adjustable guy wire rigging can significantly control boom tip deflection. The arm opposite the direction of deflection provides the greatest correction — a single arm at 10.8N reduced deflection by up to 63%. Differential tension across all three arms demonstrated 2D positioning control. Natural deflections occurred almost entirely in the expected Y-axis (in-plane with the boom seams), and measured ~5% of deployed length under dead load — unexpectedly exceeding the 1% manufacturing specification.

Designed a biomimetic soft robotic heart model that replicates native cardiac motion using pneumatic actuators embedded in 3D-printed, patient-specific ventricle geometries. The project aimed to create a physiologically accurate total artificial heart for surgical training, device testing, and hemodynamic research.

Anatomical heart CAD model derived from patient CT scans, showing exterior surface detail and vessel connections.

Approach

Starting from patient CT scans, the heart ventricle geometry was segmented and modified in CAD to incorporate channels for soft pneumatic actuators. The actuators were designed in two orientations — circumferential and helical — to replicate the heart’s complex wringing motion during systole. Actuators were handmade from silicone using custom 3D-printed molds.

Physical heart model with pneumatic actuators at labeled positions: circumferential bands (1A, 1B), twisting (2), and longitudinal compression (3) mimicking native myocardial motion.

Left ventricle actuator design: circumferential actuators provide radial compression while helical actuators generate the twisting motion characteristic of natural cardiac contraction.

CAD model of the left ventricle with marked circumferential and longitudinal actuator placement zones.

Heart Model & Fabrication

The anatomical heart geometry was derived from medical imaging data, preserving realistic chamber proportions, wall thicknesses, and vessel connections. Custom flat actuator molds were designed in Fusion 360 and 3D-printed to cast multi-channel pneumatic bladders from Dragon Skin silicone.

Transparent view revealing internal ventricular chambers and wall structure used to guide actuator channel placement.

Flat actuator mold design (v6) for casting multi-channel silicone pneumatic actuators with integrated air inlet.

Analyzed numerous configurations of people in classrooms and offices using computational fluid dynamics (CFD) to determine the best strategies for ventilating buildings and providing clean air without wasting energy. The work modeled SARS-CoV-2 aerosol spread patterns across different room geometries, ventilation systems, window configurations, and occupant arrangements.

3D CFD temperature field of an occupied classroom with ceiling-mounted supply diffusers, showing thermal plumes rising from occupants and airflow interaction with the HVAC system.

Simulation Methodology

Used ANSYS Fluent to model transient airflow, temperature distribution, and contaminant transport in classroom and office environments. Simulated human occupants as heat-generating bodies with aerosol emission from breathing zones, testing variables including window placement, heating vs. cooling modes, and natural vs. mechanical ventilation.

Airflow pathlines colored by residence time, showing how ventilation configuration affects aerosol transport and stagnation zones within the occupied space.

Temperature field visualization showing thermal stratification and airflow patterns for an alternative ventilation configuration.

Results

Compared local contaminant concentrations across 14 different room configurations to identify which ventilation strategies most effectively reduced aerosol exposure at individual occupant positions while maintaining energy efficiency.

Comparison of local CO concentration (normalized by well-mixed value) across 14 ventilation configurations, showing significant variation in exposure risk depending on window placement, thermal conditions, and occupant arrangement.

Room-scale temperature contour showing overall thermal distribution for baseline heating configuration with cool windows.