Pedro Zambrano-Saucedo

Pedro Zambrano-Saucedo

Electrical Engineering / embedded systems + PCB design

I build embedded hardware end to end. Schematic capture and board layout in KiCad, firmware in ESP-IDF, and the mechanical parts in Fusion 360 and on my own 3D printer.

Right now: rev B of a custom ESP32 development board, redesigned after I caught three faults in rev A before it went to fabrication. I'm looking for a hardware, PCB, or embedded internship for summer 2027.

At a glance
Program
BS EE, UTEP
Graduating
May 2029
GPA
3.76 / 4.00
Focus
Embedded · PCB design
Design tools
KiCad, Fusion 360
Firmware
ESP-IDF, C / C++
Languages
English, Spanish

Projects

Hardware I designed, built, and debugged myself. Each one lists what it is, what broke, and what I changed.

ESP32 Development Board

Rev B, ready to order

Custom PCB / KiCad | schematic to routed two-layer layout

An ESP32-WROOM-32 development board designed from a blank schematic to a routed layout. A USB-C receptacle with 5.1 kΩ on both CC pins, an AMS1117-3.3 regulator for the 3V3 rail, and a CP2102N USB-UART bridge with an auto-reset circuit built from two MMBT3904 transistors, so the board flashes straight from the IDE with no jumper and no external programmer. Two 20-pin headers on a 2.54 mm pitch, 25.4 mm apart, so it fits a breadboard.

Rev A got as far as a JLCPCB order before a net-by-net review of the board file showed it would not have worked, and I cancelled it. The regulator's input sat on a net that only reached the CP2102N's VBUS sense pin, so the 3.3 V rail would never have come up. A PWR_FLAG on that net was silencing the ERC error that should have caught it. Both buttons pulled EN and IO0 low through reverse-biased LEDs, so neither did anything. The auto-reset transistors were driven through capacitors instead of resistors.

Rev B fixes all three. It also replaces the USB-C plug with a receptacle, the 1 mm headers with 2.54 mm ones (the 1 mm pads were 0.15 mm apart, far below the fab's limit), and every 0201 part with 0603 for hand soldering. I routed it myself: 0 DRC errors, every exception documented, and every pad checked against the schematic netlist. Next is the order and bring-up: 3.3 V with the module unfitted first, then USB enumeration, then the reset waveforms on a scope.

Stack / KiCad | ESP32-WROOM-32 | CP2102N | AMS1117-3.3 | MMBT3904 | USB-C

Handheld Game Console

Working

Embedded firmware / ESP32 | breadboard to 3D printed enclosure

A handheld console running three games I wrote from scratch. Flappy Bird, Snake, and Tetris. I built it on ESP-IDF rather than the Arduino framework on purpose: it forced me to learn the real toolchain (CMake/Ninja, component structure, menuconfig, the idf.py flash/monitor loop) instead of dropping in libraries I didn't understand.

The firmware side taught me more than I expected. Fixed-step game timing so speed doesn't depend on frame rate, gravity and collision logic, sprite and background drawing, and input debouncing that survives fast repeated presses.

The enclosure came out of testing it. The joystick breakout has solder tails on the back that dug into my hand every time I picked it up, so I printed a cover for it before the console body existed. The body followed the same logic: the USB cable leaves through the side because routing it internally added enough thickness to make it awkward in a pocket.

Stack / ESP32 | ESP-IDF | C | CMake/Ninja | I²C OLED | Fusion 360 | FDM printing

Competition Robot (Three-Generation Redesign)

Competed

FIRST Tech Challenge, team 15450 | 2024-2025 season | Fusion 360 + printed parts

One season, one problem, three robots. The game asked us to collect smooth rectangular samples and score them, then hang the robot in the last thirty seconds. Every redesign came from a specific failure on the field.

Gen 1 couldn't physically reach into the sample area. Gen 2 added a second, deliberately thinner claw below the main one, sized to slip into the area without disturbing the samples around the target, mounted on a linear extension, the reach was still short and the extension flexed under its own length. The final robot replaced it with a slide of the same type we used for hanging, which was stiff enough to carry the reach, and added a triangular truss so the arm stopped bending and jamming at full extension. The truss was my design.

The grip was a separate problem. The samples were smooth enough that they slipped out of the claw mid-cycle, and a slip cost 5-10 seconds to recover on a cycle that ran 7-11 seconds. A dropped sample roughly doubled what that cycle cost us. Thickening the jaws didn't fix it, so I moved to casting two-part RTV silicone grip tips in a mold I designed.

Anchoring took three tries. The first jaw had a grid of small holes meant to key the silicone in place, the holes were too fine for the material to flow through, so the tip peeled off under load. The version that worked replaced them with a single large through-hole at the tip, enough for the cast silicone to pass through and lock mechanically rather than relying on adhesion. Over the rest of the season one tip came off once, and I replaced another when it lost tack before the final event.

The material choices ran in both directions on purpose. Silicone doesn't bond to PETG, so the mold is printed in PETG and the cast part releases cleanly, and that same property is why I reprinted the claw itself in PLA. I needed the tip to stay stuck to one part and let go of the other, so the two had to be different materials.

Fixing the grip then caused a second failure. Claws sticky enough to hold a sample also stuck to the field floor, not enough to stop the robot, but enough to break bottom claws and several servos. A teammate redesigned the bottom claw geometry off that finding, we bench-tested it and it held samples without any silicone at all, so I never cut a second mold for it. The right answer there was to stop adding material and change the shape.

After the grip fix we averaged about 12 samples per match and peaked at 17.

What was mine: I built the physical robot and rebuilt it in Fusion 360, reprinted the claw in PETG after the original material kept cracking in service, proposed and designed the triangular support structure, and designed and molded the grip tips. The robots were a team effort, those pieces were mine.

Stack / Fusion 360 | FDM printing | PETG | two-part casting mold | goBILDA hardware

Titan Science Payload (Concept Design)

Presented to NASA judges

Systems engineering / UAH InSPIRESS | Chief Engineer, 11-person team

UAH and NASA set high school teams the problem of designing an autonomous science payload for a Titan mission. The constraints were real and hard: fit inside 44 × 24 × 28 cm, stay under 10 kg, and survive −179 °C at 1.5 bar in 1.352 m/s² of gravity. Everything below was driven by those four numbers.

I was Chief Engineer. What I actually did was set the mission architecture and defend it: I pushed for two separate exploration paths. One staying with the orbiter, one sinking into Ligeia Mare. When most of the team thought a single payload was all we could manage. I chose the instrument set against the science objective, and I reviewed the design and analysis work the rest of the team produced rather than producing it myself. Three concepts were scored in a weighted trade study, two were selected and merged so each exploration path had a payload suited to it.

Selections were made on weighted figures of merit, not preference. The science objective came out as Life Supporting Elements (216 against 210 and 81). The shell material came out as Ti-6Al-4V (Grade 5) over two alternatives (126 against 114 and 72), scored mostly on density and durability. The payload carries a mass spectrometer, hyperspectral imager, thermocouples and an IMU, with the imager placed at an unobstructed aperture because the traceability matrix required a clear view.

Final mass budget by function, against a 10 kg limit.
FunctionComponentsMass (kg)
DeployParachute0.14
Measure1 HSI, 2 mass spectrometers, 2 thermocouples, 2 IMUs2.98
Collect data2 on-board computers0.11
Provide power2 batteries0.90
Send data2 transceivers, 2 antennas0.82
House payloadBoth payload shells3.27
TotalLimit: 10.008.23

The analysis ran in four stages. Orbital velocity at the deployment altitude worked out to 1584 m/s. The payload leaves the deployment tube at 18.4 m/s under 285 m/s². The descent settles to 2.2 m/s vertical, dropping to 0.77 m/s terminal under the parachute at 2.51 g of load. At the lake we checked buoyancy rather than assuming: 10.8 N of weight against 1.32 N of buoyant force in a methane lake of 450-600 kg/m³, so the payload sinks as the mission requires.

What was mine: the two-path mission architecture, the instrument selection, and review of the team's design and calculations. The CAD was the design team's and the numbers were the data analysts'. My job was deciding what to build and checking that what came back held up. The team took first place in Outreach at the UAH competition.

Stack / Weighted trade studies | science traceability matrix | mass budget | requirements compliance | risk analysis

Automatic parking gate with 3D printed arm and sensor mount

Automatic Parking Gate

Complete

Mechatronics / sensors + 3D printed mechanism

A sensor-triggered gate that detects an approaching vehicle and raises a barrier arm. I designed and printed the mechanical half myself. The breadboard mount and the pivoting gate arm, which is where the real problem was.

The first prints didn't fit. Printed dimensions came out under nominal because of the printer's own tolerance, so the pivot bound and the mount wouldn't seat. I measured the error, adjusted the clearances in CAD, and reprinted until the fit was right. That print-measure-adjust loop is the part I'd point to in an interview, not the sensor code.

Stack / Arduino · C++ · Fusion 360 · FDM printing

Ladder diagram of the traffic light control program in CodeSys

PLC Traffic Light Controller

In progress

Industrial control / CodeSys

A traffic light sequence written in Ladder Diagram in CodeSys, structured as an explicit state machine rather than a chain of timers. Each phase owns a latched step bit set and reset with S/R coils, a TON holds the phase for its interval, and the lamp coils are driven from the step bits alone, so an output can never disagree with the state the program thinks it's in. The amber flash is counted by a CTU rather than timed, which makes the flash count one value to change instead of a rewrite.

Part of a self-directed curriculum I'm working through alongside coursework, after MATLAB/Simulink and RC/RLC analysis. I picked ladder logic because it's how control systems are actually written on a plant floor, and almost nobody learns it in the first two years of an EE degree. The full cycle runs in simulation. The next revision adds a stop input that clears every step bit, because a sequence you can't halt isn't a controller.

Stack / CodeSys · Ladder Diagram · IEC 61131-3

Algorithmic Trading Systems

Tested, not deployed

Software / MQL5 | EUR/USD H1 parameter sweep

Five expert advisors written in MQL5 for MetaTrader 5 using moving-average and RSI entry rules on AUD/USD and EUR/USD, with stop-loss and take-profit brackets and position sizing enforced in code. The sweep below is the moving-average expert on EUR/USD H1, 1 Jul 2024 to 1 Jan 2025, 3,133 bars, 5.02 M ticks, 99% history quality, every-tick modelling, 10,000 USD deposit at 1:100. I ran 14 parameter variants against that identical history and logged net profit, profit factor, maximum equity drawdown, and Sharpe ratio for every one.

All 14 runs, sorted by net profit. Nothing excluded.
RunNetProfit factorMax equity DDSharpe
01+1,455.661.1248.44%0.49
02+1,441.241.1153.21%0.43
03+1,420.111.1057.55%0.38
04+1,389.011.1443.25%0.54
05+1,309.791.0861.51%0.31
06+1,187.421.0765.09%0.26
07  +500.521.0543.22%0.21
08  −115.900.9937.48%−0.06
09−1,156.550.7830.56%−0.79
10−1,436.430.6718.75%−1.61
11−1,517.790.7526.67%−1.14
12−2,185.750.6527.51%−1.70
13−3,182.420.7845.86%−2.03
14−4,456.840.6649.77%−2.95

Seven variants finished profitable and seven didn't. The best profit factor was 1.14 over 43 trades, which is nowhere near enough trades to separate an edge from luck, and the best run needed a 43% equity drawdown to return 13.9%. More telling: across the sweep, gross profit scaled with drawdown while the trade count stayed fixed at 43, the variants that made more money weren't finding a better signal, they were just sizing up. My conclusion was that this strategy family has no edge worth trading. I'd rather publish that than a return number I can't defend.

Two limits worth stating before anyone else does: the tester ran with zero-latency ideal execution, so no slippage or requotes are modelled, and six months of H1 on a single pair is one market regime, not a sample of many. Both push the results optimistic. Neither changes the conclusion, a profit factor of 1.14 doesn't survive either correction.

The engineering held up better than the strategy. An EA runs on every tick, so entries have to be idempotent, orders reconciled against what the broker actually filled, and behaviour identical between the tester and a live account.

Stack / MQL5 | MetaTrader 5 | Strategy Tester | EUR/USD H1 | 14-variant sweep

Arduino fundamentals

The exercises I started on digital I/O, sensor reads, LCD output. Coursework-level, kept here for the record rather than as portfolio work.

Fusion 360 practice

Parts I modelled and printed on my own machine. Most exist because a project needed a bracket, a mount, or an enclosure that didn't exist yet.

Printed breadboard carrier and gate arm laid out on the bench

Parking gate arm & mount

Fusion 360 / FDM

Pivoting barrier arm and a breadboard carrier. Reworked twice after measuring how far the printed dimensions fell from nominal, then adjusted clearances to suit.

Notes / tolerance fit · printed in PLA

Two-part Fusion 360 mold for casting claw grip tips

Claw grip-tip mold

Fusion 360 / tooling

Two-part split mold with locating pins, a pour channel, and drafted walls so the cast tip releases cleanly. Built to solve a grip problem on the competition robot above.

Notes / PETG mold | two-part RTV silicone | PLA claw

Geared bottom claw mechanism on its servo mount

Bottom claw mechanism

Fusion 360 / competition robotics

Geared jaw pair on a servo mount, sized around the game element so the opening clears it without disturbing neighbouring pieces. Revised after the grip fix started breaking servos.

Notes / printed jaws | geared linkage

Skills

Listed at the level I can defend in an interview.

Hardware design

  • PCB design — KiCad, schematic to routed layout
  • Circuit analysis — RC / RLC, DC networks
  • Soldering — through-hole and SMD rework
  • Prototyping — breadboard to protoboard

Firmware & software

  • C / C++ — ESP-IDF, Arduino
  • Python
  • MATLAB / Simulink
  • MQL5
  • Git — branching, remotes, Pages

CAD & fabrication

  • Fusion 360 — parametric modelling
  • 3D printing — FDM, slicing, tolerance tuning
  • Hardware repair — board-level diagnosis

Instruments & systems

  • Oscilloscope, multimeter, bench supply
  • CodeSys — ladder logic (learning)
  • Linux / bash
  • ESP32 peripherals — GPIO, UART, I²C, ESP-NOW

Experience

Paid work and the competition roles where I led a team.

Jul 2026 — present

Technology Support

University of Texas at El Paso

  • Diagnosed and repaired OS-level failures on 110+ laptops under departmental turnaround deadlines.
  • Root-caused BIOS and system-level faults on 80+ units, isolating hardware vs. firmware failure points before applying targeted fixes rather than blanket reimaging.
  • Disassembled and reassembled 100+ systems for component-level repair and parts recovery.
Aug 2021 — Jun 2025

Coach and 3D Designer

FIRST Tech Challenge Robotics

  • Competed in 20+ matches across multiple seasons, advancing to state and premier levels.
  • Coached a 15-person team and designed 10+ 3D-printed prototype robots.
  • Raised $5,000+ per season in sponsorship.
Jan 2024 — Apr 2025

Chief Engineer and 3D Designer

UAH InSPIRESS

  • Led two competition engineering teams, qualifying for the Alabama competition.
  • Presented a prototype simulation to NASA judges. Won first place in Outreach.
Jun 2024 — Aug 2025

Math Tutor

DaVinci School for Science and the Arts

Adapted explanations to each student's level across a range of ability. Helped 110+ students passed their STAAR exams.

Education

2025 — 2029 (expected)

B.S. Electrical Engineering

University of Texas at El Paso

GPA 3.76 / 4.00.

Ongoing

Self-directed engineering study

Independent

MATLAB/Simulink coursework, RC and RLC circuit analysis, and IEC 61131-3 ladder logic in CodeSys. Worked through alongside my degree to get to applied hardware sooner.

Involvement

Member

AESS — Aerospace and Electronic Systems Society, UTEP

Technical workshops in soldering and PCB design.

Member

IEEE UTEP Student Branch

Technical and networking events.

Member

SHPE / MAES, UTEP

Technical workshops and professional development events.

Contact

Open to hardware, PCB, and embedded internships, and happy to talk through any of the projects above in detail.