Directions without taking your eyes off the road

Riders who follow a map on a phone look away from the road. In one weekend at CalHacks 13.0, we built a retrofit helmet that cues directions and signals to the people around the rider.

• Directions by buzzerThree buzzers, left, front and right, cue which way to turn.
• Blinkers and brake lightThe blinkers switch on automatically from the GPS signal, and a brake light shows at the back.
• Voice navigationA button and microphone on the helmet take a spoken destination, such as “navigate me to…”.
• A companion appA map and destination search, with the trips kept in a database.

My part was the two housings, the app’s UI/UX and the demo video, on the next tabs.

Housings for flat electronics on a curved helmet

I designed and 3D printed the helmet’s two housings: one for the power bank and one for the microcontroller circuit. The central design question was how to mount flat electronics on a curved helmet, so I designed planar boxes and chose their position and angle from caliper measurements.

• Measured firstI measured the helmet and the electronics with calipers, and worked out the 50° angle between the boxes from those measurements.
• Mounted at the backI weighed the top against the back and chose the back, because I expected it to disturb the airflow less.
• Two boxes, with slotsA power bank box and an electronics box. The electronics box has one slot for the cables from the power bank and one for the cables to the buzzers, microphone, LEDs and button.
• Two versionsVersion 1 of the power bank box and version 2 of the electronics box ended up on the helmet. Version 2 added mounting eyelets and was printed on Bambu Lab printers in a 5-hour window.

The electronics box was sized for an Arduino that joined the circuit late, because the blinkers ran at 5 V and the ESP32 at 3.3 V.

From paper sketch to app screens

I designed the app’s UI/UX: a paper sketch and feature list first, then the pages, palette, icons and flow in Figma. I refined its animations by vibe coding.

The sketches

The screens

• Sketch and feature listI started from a paper sketch of the app and a list of the features.
• Pages and flowThe pages and the transitions between them, including the GO transition into the ride screen.
• Palette and iconsA dark screen with a yellow accent, and icons for the rider’s bicycle or scooter.
• AnimationsI refined the animations by vibe coding, including the transition into the ride screen, to polish the interaction.

The demo video, and what I learned

What I learned

• Retrofit means fitting flat electronics to a curved helmetThe helmet surface is curved, but the circuit boards are flat. A curved housing would follow the helmet and still not hold the electronics, so each housing is a flat-faced box. The fit then comes from where the boxes sit and at what angle: I measured the helmet with calipers and chose the back of the helmet and a 50° angle between the two boxes, so that two flat housings sat on the curve.
• Have the housing ready before the electronics are solderedThe housing and the circuit have to be agreed together, covering sizes, slots and cable routes, and the housing needs to be ready before the electronics are soldered. Once a circuit is soldered, changing it means taking it apart again, which costs time during a one-weekend build. Next time I will finish and test-fit the housing first, so that the final circuit goes into a housing that is already made.

Where the helmet goes next

The demo helmet shows the concept working. These are the steps that take it from a weekend prototype towards a helmet riders could wear every day.

1. 1Real vibrationMotors for the cues
2. 2Third housingBuilt around the final circuit
3. 3Rider testingReal routes, real riders
4. 4A smarter helmetCrash detection, solar, soft electronics

• Real vibration for directionsThe demo helmet cues left, front and right with three buzzers. The next version adds vibration motors, so a rider can feel the cue as well as hear it.
• A third housing, designed with the final circuitWith the electronics fixed first, I would design both boxes together, keep them to as few parts as possible and keep the 50° fit to the helmet’s curve.
• Test with ridersTake the helmet out with riders on real routes and see what they notice, how the cues feel and how the helmet wears over a longer ride.
• Lighter, softer electronicsIntegrate the electronics into breathable fabric, so the helmet carries less hardware on the shell.
• Crash detectionUse the data the helmet already collects to detect a crash and call an emergency contact automatically.
• Power and preventionA solar panel to keep the batteries charged on long rides and, much further ahead, a camera for crash prevention.

Automating how a carbon-capture material is deposited

MOFs (metal-organic frameworks) are porous materials that can capture carbon dioxide. In the Design for Emerging and Nanoscale Manufacturing lab at Berkeley, our team of three is developing a rig that deposits a MOF automatically.

• My roleAssembly Lead. I own how the parts fit together and I am building the compact assembly in Autodesk Fusion, with 3D-printed mounts.
• StatusIn progress. The design is still being developed, and its details are confidential for now.

Helping someone hold and position a drink with less assistance

Holding a drink steadily takes grip, coordination and control. When those fade, drinking depends on someone else. The project explores how people with limited hand strength or motor control could drink with less help, while keeping their dignity.

• The needTo drink with less help, without someone holding the cup.
• DignityIt has to feel right for an adult, not like a child’s product.
• The conceptNext: brainstorm, then choose a concept as a team.

Why we chose drinking

We interviewed a patient recovering from surgery and a caregiver, to find gaps in daily life that a design could fill or improve. Most of what points to drinking comes from the caregiver’s account.

• Spills and dropsThe person the caregiver looks after struggles to hold a cup, spills drinks, and sometimes drops the glass when she falls asleep holding it.
• Not a sippy cupAn adult sippy cup feels childish, so it would not be used. The less childlike a device feels, the more likely it is to be used.
• The patient named other strugglesGetting out of bed, dressing and finding a walker sturdy enough for the street. He did not name drinking.
• Fine movements are hardRemotes and phones are difficult. Voice-controlled devices work better.

How we chose

• Other directionsWalkers and hospital beds, which came from the patient interview.
• Why drinkingIt was specific, achievable in about 10 weeks and within the team budget.
• What existsAdult sippy cups and an assisted-gravity cup, which addresses oral-motor limits. We are still looking at the gaps.

I took the notes in the patient interview.

The problem is chosen; the concept is not

1. 1InterviewsDone
2. 2Framing the problemDone
3. 3Concepts and researchNext
4. 4PrototypesLater

The safety boundary

• Easier is not saferSome people have trouble swallowing, staying alert or controlling movement, so a device that is easier to handle is not automatically safe for independent drinking.
• Not for lying flatWe will not promote drinking in a flat position.
• Who it is forWe will define the intended users and write safety guidance as part of the design and testing.

A newsletter designed to be read in two to three minutes

• Mobile firstDesigned for the phone screen before anything else.
• On brandFollows the BlueRobins identity and design guidelines.
• Every section looks differentSo the issue stays varied from top to bottom.
• Fun, not childishPlayful enough to enjoy, grown-up enough for a parent.

Who to write for, and what goes in

For every candidate section I decided whether it would run every issue, rotate, or stay out. Five made it in.

From a whiteboard to a phone screen

1. 1BrainstormSections kept, rotated or dropped
2. 2SketchEach section drawn as a block
3. 3WireframeMade with Claude and Figma
4. 4PolishSmall design changes in VS Code

• Whiteboard sketch
• Wireframe variants

• Chose the sectionsBrainstormed the options and decided which ran every issue.
• Sketched themEach section drawn as a different block.
• Proposed the quizMy main idea: a question for the family, always on.
• Mocked up the designA wireframe made with Claude and Figma, at phone width, and the demo newsletter.
• Presented the videoI was the voice and face of the pitch.

Scroll the phone, or jump to a section

• HeaderThe brand header, locked on top of every issue.
• 1 Project highlightOne mentor-led project: a photo, a title and a short description.
• 2 Featured mentorA mentor card, so a parent knows who guides the students.
• 3 TestimonialA short quote from a parent.
• 4 Success storyA student's project, in a few sentences.
• 5 QuizA question for the family with four answers. I proposed this section, and it is always on.

The photos and quotes are sample content. The layout is what we designed.

Winner of the Launch-a-thon

Turning a page takes a pinch, a grip and steady pressure

3 hands-free ideas, 1 Pugh matrix

I introduced the double diamond to the team and set up the Pugh matrix to converge, scoring seven criteria from ease of use and safety to cost and ergonomics. I proposed two of the three starting ideas, the hair brush and the ponytail tier.

• The gap we sawA printed turner's arm blocked the page and its friction marked it. "Book Time" worked but was dear and built for magazines.

The base, the tilting back and how it is made

I modelled the body that holds the book, then wrote down how it is made and put together. The back tilts up to 70°. In the model the angle is set by hand, and the motorised adjustment the team planned is still to be modelled.

12345

1. 1Tilting back. It holds the book and is designed to tilt up to 70°. The angle is set by hand in the model.
2. 2Tilter link and hinge. They carry the back and hold its angle.
3. 3Recess in the base top. The tilter seats in it, which is why the back and tilter go in first (assembly step 1).
4. 4Base housing. Made as two halves joined together (step 2), which should make it easier to mould.
5. 5On/off button. It is colour-coded and also carries a text label, so it does not rely on colour alone.

• Section view (mine)
• Exploded view (mine)

Materials, processes and assembly

• Calm to look atI chose neutral colours and a different colour for the moving parts, so they are easy to tell apart.
• Labelled for colour-blind usersThe on/off button carries a label as well as a colour, so it does not rely on colour alone.
• Easy to live withA wrist cushion, adjustable pedals, a box with handles and a strap, and different colours for the left and right buttons.

The pad lifts the page, the flipper carries it across

Team design. The video loads from YouTube only when you press play, or watch it on YouTube.

• LiftAn electroadhesive pad, copper electrodes under a film, pulls one page to the pad when it is switched on.
• CarryA flipper slides under the gap and carries the page across. It works in both directions.
• TriggerAny single-action switch: hand buttons, foot pedals or a mouth-pressure actuator.

12345

1. 1Tilting back (mine). It holds the book.
2. 2Base (mine). It carries the back and the module at its side.
3. 3Separator arm (team). Its electroadhesive pad lifts one page.
4. 4Flipper (team). It slides under the gap and carries the page across.
5. 5Servo (team). One of the servos that move the arm and the flipper.

• Separator and flipper (team design)

The flipping module is replaceable for repairs and upgrades. A master sketch checked how the parts interact before modelling.

2nd of five finalists, with a prototype as the next step

• ResultSecond place at The CAD Challenge 2026, one of five finalists. The prize included an internship interview with Hestus.
• Known limitsThe pad surface wears, a wet page weakens the hold, and a bigger battery means a heavier base.
• NextBuild a prototype to check that the electromechanical system works.

Sources and assumptions. The 19.9 million figure is from the US Census Bureau (2012). The $2,000 to $3,000 price of automatic turners is quoted in our report. Pugh scores come from the team's matrix against the hair brush as reference. Book size and weight limits are our design targets, not measurements.

Pharmacies close early, so working people miss their medicine

Four choices, each settled with a Pugh matrix

Concept generation and selection was my section of the report. For each choice I scored the options against a reference.

Each criterion is scored from −2 to +2 against a reference option, and the scores are added up.

Sizing, cooling, power cuts and materials, by hand

The technical feasibility was my section of the report: how big the locker is, how it stays cold, how it survives a power cut and what it is made of.

Sizing: 120 drawers a locker

The market inputs come from the team's market analysis. I used them to size the drawers.

Keeping it cold: 13 W leaks in

• Held at 5 °CThe middle of the 2 to 8 °C range for medicine. Shop assumed at 25 °C.
• R290 refrigerantZero ozone depletion and very low global warming impact.
• About 3 p a dayCompressor electricity at 27.69 p per kWh.

Power cuts: three hours of backup

• 3 hours of backupUK power cuts average about 2.15 hours, so there is a margin.
• 12 V LiFePO4Chosen over sealed lead acid: its capacity holds up at higher discharge rates.
• Denser and longer-livedDenser than lead acid, which matters in the base, and a much longer cycle life.

Materials and making

• 304 stainless steelRust resistant, hygienic and secure. Laminate was dearer and heavier.
• 1.52 mm and 1.22 mmChassis and drawer-wall thickness, from standard sheet gauges.
• Clear acrylic bottomsYou can see nothing is left in an upper drawer.
• TIG weldingPrecise on thin sheet. Strength against break-ins is a design target for the next stage.
• Built to be reusedRecyclable steel, a target 20-year life and swappable parts.

A locker inside the supermarket you already visit

The concept is the team's. A patient gets a one-time code when their prescription is ready, and collects it while shopping.

1. 1Prescription readyThe patient gets a one-time QR code
2. 2Go shoppingLockers sit inside supermarkets
3. 3Scan the codeOr type a backup number
4. 4One drawer opensA spring latch releases that drawer only

1234

1. 102 ambient drawersSized to the largest UK medicine carton
2. Screen, QR scanner, keypadSpeaker, camera and help button too
3. 18 refrigerated drawers, 2 to 8 °CLowest, so everyone can reach them
4. Compressor and 12 V backup batteryIn the base, with a power-cut alarm
5. About 161 cm tallAn average UK woman's height

A concept study, with the cooling and battery sized by hand

Sources and assumptions. Pharmacy statistics are from the Company Chemists' Association and the Royal Pharmaceutical Society, as cited in the team report. UK power-cut length is a UK-wide average, not a Bristol figure. Heat leak, compressor duty and battery sizing are my hand calculations from supplier datasheets.

It picks the outfit, so you don’t have to look at the sky

A website that suggests what to wear from the weather, how warm you run and the clothes you actually own, and tells you what to pack for a trip.

• YouHow warm you run and your skin tone, set once.
• WardrobeOnly the clothes you own, so nothing is suggested that you don’t have.
• TodayThe forecast, three looks for the day, and a look for each time you are out.
• TripsA packing list from the real weather your dates had in recent years.

Two answers that change every suggestion

I decided what the website should know about you before it suggests anything: how you feel the cold, and what you own.

• How you usually feelCold, average or hot. It shifts every look about two degrees warmer or cooler, because some people feel the cold at 18 degrees and others are warm at 12.
• Skin toneUsed for sun advice. Fairer skin gets a firmer sunscreen reminder on high-UV days.

• 35 pieces, 6 groupsTops, layers, coats and jackets, bottoms, shoes and extras, each drawn as a flat illustration.
• Tap what you don’t ownEverything starts selected. Untick a piece and it is never suggested. Here the skirt and the dress are unticked, so you are never told to wear a dress you don’t own.

I decided what it does; AI wrote the code

• What I decidedThe four tabs, what the website asks you, how warm you run shifts every look, which clothes count as yours, and how a day or a trip turns into outfits.
• What AI didIt wrote the code that makes the website behave that way.

Today and Trips, as the website shows them

• The forecastTemperature, how it feels, rain, wind and UV, with a measuring tape from the day’s coldest to its warmest.
• Three looks, A, B and COutfits built only from the clothes you own, with a tip such as taking the light sweater off once it warms up.
• Times you’re outsideAdd the hours you will be out and get a look for each. A cool morning gets a jacket and jeans, a mild afternoon gets light trousers.

1. 1DestinationWhere you are going
2. 2DatesWhen you leave and come back
3. 3Real weatherWhat those dates were like in recent years
4. 4Packing listCounts per piece, to tick off

• Weather those dates hadWarmest and coldest feel, sunny days and days it may rain, from real weather on the same dates in recent years.
• Counts, not just a listIt says how many to take, such as five t-shirts, and why, such as shorts for about three of four days.

A working website, shown with sample data

Good hands cost a fortune, cheap hands barely grip

About 5,200 upper-limb amputations happen in the UK each year. Myoelectric hands cost £25,000 to £60,000. Printed ones are affordable, but weak.

Show the finger, labelled

Distal phalanxthe fingertip

Middle phalanx

Proximal phalanx

Metacarpalthe palm bone

DIP jointdistal interphalangeal

PIP jointproximal interphalangeal

MCP jointmetacarpophalangeal

Cable holesthe tendon threads through these

Elastic extensors first, then springs

To open the finger again we tried elastic extensors, in two printed materials, before moving to torsion springs.

Elastic resin 50A

Too soft. Six prints tore, supports would not come off, and the simulation would not converge.

✗ Dropped

TPU 90A

Stiffer, and it printed cleanly for the ring and middle fingers. Inserts deformed, but the material was worth simulating.

✓ Simulated next

I simulated whether TPU, then springs, could open the finger

My streams were printing and testing the passive extensors, redesigning the MCP extensor, and simulating two torsion spring designs in Siemens NX.

Can TPU open the finger after it grips a tennis ball?

Ring finger, 66 mm tennis ball at the optimal grasp. Flexible-body extensors (1.6 to 1.9 mm tetrahedra), springs driving the motion, 1.2 s simulated.

Redesign: cut the links that held tension at rest

1Original · start

2Original · end

3Cut · start

4Cut · end

1. 1Tension held at rest. The ringed links are under load before the finger moves (up to 0.62 MPa), which fights the grip.
2. 2Slack at the end. Nothing is left pulling, so it cannot open the finger.
3. 3Fixed at rest. With the links cut there is no tension before the grip.
4. 4Slack at the end. The redesign fixed the initial tension, and the slack at the end led me to recommend extensors that follow the bones’ curvature more closely.

Stage	Beginning	Middle	End
Should carry	No tension	Some tension	Some tension
Redesign carried	No tension ✓	Some tension ✓	No tension ✗

Fixed at rest, still slack at the end. I recommended extensors that follow the bones’ curvature more closely, and moved on to springs.

Two torsion spring designs, simulated on one finger

One torsion spring at the DIP joint and two at the PIP. Design 1 is preloaded; design 2 scales the bones 1.5 times, so it is bigger than a human hand, because the 5 mm bearings and the springs had to fit.

The team's hand: one cable, under 400 g

The group built a prototype hand. Everything on this tab is the team's result, not only my part.

• One cableA single tendon closes all five fingers.
• Under 400 gThe weight of a biological hand, which was our ceiling, and the prototype stayed under it.
• Two graspsThe prototype demonstrated power and pinch grasps.

Both spring designs open the finger in simulation; grip is the next test

Sources and assumptions. Amputation numbers, hand prices, grip forces and abandonment rates come from the sources cited in the group report. Stresses and forces are Siemens NX simulation outputs for one finger, not measurements. The 2,420 MPa figure is the spring supplier’s ultimate tensile strength, used as a reference, not a yield limit.

Hold 19 °C inside a thin tent on a −2 °C night

A box, two sleepers and 27 stated assumptions

I modelled the tent as a 2.3 × 1.3 × 0.75 m box with the two sleepers as the only source, about 150 W between them, and worked out each way heat leaves by hand.

Conduction, convection and radiation use Grashof, Reynolds, Prandtl and Nusselt number correlations. Most of the 27 assumptions lean to the worst case, and my report marks each as an over- or underestimate, so the result is a conservative estimate.

The tent loses 4.0 kW, mostly on the wind

The tent loses 4.0 kW. The two sleepers give back about 150 W, so heating has to make up 3.9 kW, 26 times what they give.

Three passive fixes cut the loss by 82%

One cold night, two average people

Sources and assumptions. The net 3.9 kW is the 4.0 kW lost minus about 150 W from the sleepers (the report gives 149 W). The 1.37 kW comparison is a published army-tent study scaled to two people. The tent model, the 27 assumptions and every loss figure are my hand calculations.

Will a welded lug survive lifting a 12-tonne vessel?

Six lugs share the weight of a pressure vessel, and each is welded to a plate that is welded to the vessel. I checked one lug for strength, for a pull that is off-axis, and for fatigue, then proposed an inspection interval.

One element by hand, checked against Abaqus

A six-node triangle solved by hand, checked against a converged Abaqus mesh (C3D10 elements, 4 mm).

• Plane stressThe hand model treats the lug as a flat sheet and ignores the fixed back plate, so it cannot capture the eye.
• LoadThe ×3 safety factor is already inside the 58.86 kN, so no further factor is applied.
• Constants from the course briefFracture toughness K_IC 60 MPa√m, geometry factor Y = 1.12, and Paris' law constants C = 10⁻¹¹ and m = 4 were given, not measured.

Two models agree, and a pull off-axis is the worst case

The two models agree closely on how far the lug moves.

0.0131 mmdeflection, by hand

0.01277 mmdeflection, Abaqus

2.5%apart

What if the cables pull off-axis?

Yield-to-stress ratio of 5.6 at the weld, 2.7 at the highest-stress point

12

1. 1The load goes through the eye. 58.86 kN per lug, safety factor included. The highest stress is here, 131.1 MPa, a yield-to-stress ratio of 2.7.
2. 2The weld joins the lug to its plate. 63.22 MPa, a ratio of 5.6. This is the stress range I used for the fatigue calculation.
3. 3The calculation that follows. A crack growing from the smallest one an inspection finds (1.46 mm) to the allowed size (4.05 mm) takes 1.77×10⁵ cycles, so I proposed inspecting every 10⁵ cycles.

Fatigue: a proposed inspection interval comes before the crack reaches its allowed size

Hand calculations and Abaqus FEA, and what comes next

Sources and assumptions. Load, material properties, K_IC and the Paris constants come from the course brief. Stresses and displacements are my hand calculation and an Abaqus CAE model. The proposed inspection interval and cycle counts come from the coursework Paris' law model and are not fatigue tests.

Pack flat on the roof, open to face the sun

We built the linkages in LEGO before choosing

I built three of the four concepts in LEGO, including the one we chose.

• 

  Concept 1 · Scissor and extending beam

  Panels pack vertically. 0 to 90°.

• 

  Concept 2 · Sliding panels

  Panels slide into slots. About 50° at most.

• 

  Concept 3 · Simple scissor Chosen

  Panels sit on the arms. 4 to 6 panels, 0 to 80°.

• 

  Concept 4 · Gear parallelogram

  Sweeps past vertical, to about 100°.

How we chose

A Pugh matrix scored each concept against criteria taken from the specification: manufacture and assembly, usability, panel performance, deployed panel area against packed volume, and speed. Concept 3 won because one scissor lifts several panels with few parts, only beams and screws, and it deploys fast and packs small.

A proposed failsafe that stops the motor, even with a button held

As team lead I wrote the specification and built the LEGO concepts. Then I designed the supervisory controller, chose its sensors and wrote the Arduino interlock code. The motor runs only while all three hold:

Two scissors, a driveshaft and three rows of panels

The final design is the team's. Two scissor mechanisms sit on the sides, joined by a driveshaft, with three rows of two panels between them.

41LEGO concept 3

123Final design

1. 1Kept: a scissor of beams that carries the panels. This is why concept 3 won, and it is still the heart of the final design.
2. 2Changed: two scissors, one on each side, with three rows of two panels between them, for rigidity and more panel area.
3. 3Added: a drive shaft joining the two sides so they move together.
4. 4Where the geometry was first tuned. LEGO let us change linkage lengths and mounting angles quickly. In the final design the lengths changed again, to point the panels more squarely at the sun.

• Lighter panelsSmaller 1.4 kg panels brought the deployed height from 3 m to 0.80 m.
• A vertical lead screwIt cut the force needed to start from 20,000 N to 143.7 N, and it is self-locking.
• Aluminium beamsAA6061-T6, with a safety factor of 5.69, within a £50 material budget.

It met all but one key requirement in the calculations

Sources and assumptions. Figures come from the group design report (March 2025). The roof-tent company is a course scenario, not a real client. The specification, the Pugh matrix and the component choices are in the report's tables.