From breadboard
to built.

Most homes have no reliable answer to “is the door actually locked?” Commercial smart locks solve it by replacing the whole deadbolt — $100–350+, a visible aesthetic change, and no coverage for windows at all. My concept: a cheap, invisible add-on that reports lock state on doors and windows without touching the existing hardware — funded and technically directed through my company, built with the student team below.

A small sensor clips onto the deadbolt's slide bolt (or a window's sash) and reads position with a reed switch and magnet — no rekeying, no visible hardware change. It reports locked or unlocked over Wi-Fi in real time to a phone.

• Electronics shrank across the project — from a full STM32 Nucleo dev board down to a coin-cell-sized ESP32-C3 for the final MVP.
• Firmware watches a reed switch on a pull-up pin and publishes a locked/unlocked event over MQTT the instant the state changes.
• A parallel reed-switch sensor was built for windows, extending the same approach to whole-entry monitoring — the gap the team's competitive teardown (August, SimpliSafe, Ring, Nest, and Saratech's own smart-sensor line) found no one else covered well.
• Housing went through five SolidWorks revisions (V1–V5), refining battery compartment, hinge, and thumbturn clearance for a clean non-invasive retrofit.

• Door-to-phone latency verified live — a state change on the deadbolt reaches the app in about a second (Fig. 06; Clip 01 for the bench bring-up).
• Modeled at $13.93 per unit against $100–350 for commercial smart locks; per-unit material cost fell from $9 (V1) to $3 (V3) as the design matured across three prototype generations.
• Business case explored a B2B path — partnering with door manufacturers (e.g. Marvin Doors) to embed the sensor at the factory, rather than selling direct to consumers.
• Took first place at its UCI design showcase, then went on to a full investor pitch — the path from student prototype to funding conversation, not just a class deliverable.
• That path got real traction, not just a slide: Styline Windows & Doors reviewed the prototype in person and expressed interest in participating, alongside early conversations with an outside investor.

A small fire can outrun a person's reaction time. I framed the alternative directly against traditional sprinkler systems — pricey, slow-responding, and hard to install — and directed the team to build a ceiling-mounted system that finds a fire's exact location on its own, aims at it, and suppresses it within 10 seconds of detection (an early pitch-deck target had been 30s), built against NFPA 750 (water-mist fire suppression).

A ceiling-mounted pod (≈10 ft coverage radius) reads the room with a thermal camera, drives a two-axis servo gimbal onto the hottest point above a fire threshold, and opens a solenoid valve to release a pressurized water mist once it's centered.

• Targeting logic centers the gimbal on the hottest pixel in the thermal frame once it crosses 50°C, then holds and fires.
• Live telemetry — temperature, fluid level, battery — streams over BLE to a companion app.
• The mechanical arm and nozzle went through three design generations across 87 CAD files before the final geometry; formal engineering test procedures target extinguishing a contained fire in under 5 seconds at ranges from 2 to 10 feet.
• A dedicated wiring diagram (Fig. 04) documents the full electronics build: dual servos, relay, solenoid, and battery-level sensing on an LM339 comparator.

• Project log records the system successfully putting out a live test fire on 1/29 — the video above is that test — ahead of that day's Critical Design Review.
• Finalized prototype completed 3/19 after a pass on space efficiency and a full mechanical drawing package.
• Full bill of materials came to $334.05 across 26 line items; a fabrication guide and safety/operation manual were written for repeatable builds.
• Pitched under their own brand, Fire-Fight Club, with the funding case resting on three points: inexpensive and easy to install, autonomous rather than manually triggered, and app-monitored end to end.

Coaches want a read on an athlete's exertion in real time, but the standard method means drawing blood and running it through a lab analyzer. The sponsor asked for a non-invasive alternative: a patch that reads lactate straight from sweat.

A wearable patch worn against the skin measures lactate in sweat through a custom potentiostat and an enzyme-functionalized membrane — built and calibrated from the electronics up, on an 18-week timeline compressed to about 10 working weeks after a delayed project approval.

• The potentiostat was validated against known hydrogen-peroxide concentrations first, bypassing the enzyme layer to confirm the electronics before trusting the chemistry.
• The enzyme membrane is hand-manufactured over a multi-day chemical process, cast onto graphite electrodes and air-dried.
• Housing iterated through a watchband-integrated capsule and a compact waistband module, wear-tested for stability, fit, and comfort.
• Requirements traceability was tracked line-by-line against pass/fail criteria — the team's own discipline is what surfaces the honest result below.
• The research phase weighed alternatives before committing to lactate — body-odor profiling and cortisol sensing were both evaluated and set aside in favor of a sweat-lactate approach.

• Electronics and mechanical subsystems passed verification outright: stable output over a full hour of simulated exercise use, independent battery operation, and a housing that stayed put and stayed comfortable through wear-testing.
• The lactate-sensing chemistry did not survive the supply chain: the enzyme sat at room temperature for 3 days in university purchasing before the team could refrigerate it, most likely denaturing it and leaving the electrode layer unable to selectively detect lactate — 3 of 4 chemistry requirements are logged as failed in the team's own traceability matrix.
• Root cause, cost, and schedule impact were all documented rather than glossed over: BOM landed at $632 against a $400 budget (a same-part replacement enzyme alone would have added $216 and another week of lead time), which the team's own retro ties directly to the compressed timeline.
• Benchmarked against every commercial lactate sensor on the market (Onasport, Pinnacle, Zimmerman Peacock, and others) to confirm the price/performance gap this device is meant to fill, and pitched for continued funding on that basis — the enzyme fix is the next item in the pipeline, not a dead end.

E-bike packs cram dense lithium-ion cells into small enclosures with little airflow — a known thermal-runaway risk. The sponsor's brief was ambitious on purpose: something genuinely new in the category, not an incremental tweak, with a hard thermal ceiling of 40°C.

A redesigned 14s3p (42-cell) enclosure pairs a battery management system with active cooling airflow, sized and verified in CFD before a single part was cut — then proven again with a thermocouple against a fan-off control.

• Airflow and fan placement were validated in Siemens StarCCM+ CFD before committing to a physical build.
• A custom cell holder carries the 14s3p pack alongside the BMS and a DC-DC converter for auxiliary power, inside an enclosure that iterated through three generations — Prototype C integrates cell holder, BMS, and fan cooling path into one case, lid, and fan-lid assembly.
• Safety envelope was set explicitly before testing began: 60A discharge ceiling, per-cell voltage window of 2.5–4.25V, and a hard 40°C thermal limit.
• Validated in a real environmental test chamber, not just on a bench — the full pack was instrumented with thermocouples and run inside a controlled-temperature enclosure (Fig. 03).

• Side-by-side proof-of-concept testing (see the chart below, and Clip 01 for the physical test) is the real evidence: with the fan off, pack temperature held flat near 50°C for 40+ minutes with no meaningful drop. With the fan on, temperature fell at an average of −0.39°C/min across trials — as fast as −0.48°C/min — pulling the pack down into the mid-30s°C, comfortably inside the 40°C ceiling.
• One trial turned the fan off partway through and watched the pack start reheating immediately — the clearest possible confirmation that the cooling path, not ambient drift, is what's doing the work.
• Delivered as a two-quarter (151A + 151B) capstone, with sponsor interviews, a pitch deck, and a poster presentation — pitched for continued funding to carry the pack past this prototype stage.