Projects

V.2.0.1

settings

VESSEL.AI COPILOT

Enterprise engineering copilot unifying 600+ vehicle features using Agentic AI, PyTorch, and Multi-Agent RAG.

#PYTORCH #RAG #AI_AGENT

Last updated: 2d ago

> Role: Tech Lead - AI&ML Engineering Platforms

Vessel.ai is Ford’s enterprise-wide engineering copilot—a single, secure point of access to everything the company knows about its vehicles. The platform gathers requirements, design drawings, safety analyses, calibration sheets, test results, warranty records, and fleet-telemetry summaries into one canonical knowledge space. Engineers can ask natural-language questions (“Which ECUs share the same LIN bus on U76?” or “Draft the ISO 26262 goals still missing tests for power-liftgate”) and receive fully sourced answers in seconds instead of spending days hunting through file shares and spreadsheets.

At its core, vessel.ai combines large-language-model reasoning with a governed search layer. All content—structured tables and free-form documents alike—is catalogued with program, confidentiality, and lifecycle metadata, ensuring that each user sees only what their role permits. When a query arrives, the system interprets intent, locates the most relevant knowledge, and assembles a concise, citation-rich response. This means a feature-systems engineer can move seamlessly from cross-feature dependency analysis to DFMEA drafting to compliance tracing without leaving the chat interface.

Early pilots cut document-search time by more than 90 percent, automatically populated DFMEA rows with high accuracy, and surfaced a dozen critical integration conflicts before design verification—saving millions in retest effort. The same foundation is now being extended outward: dealer-service assistants that surface repair procedures and parts availability; driver-facing concierge skills in FordPass and SYNC; and fleet-health dashboards that turn raw telematics into early-warning insights.

By unifying data, access control, and AI-driven reasoning in one coherent environment, vessel.ai is reshaping how Ford designs, validates, and services the vehicle—transforming decades of tribal knowledge into on-demand intelligence for every engineer, dealer, and eventually every driver.

Key Features:

Agentic AI for autonomous task execution & reasoning
Role-based access with full audit trail
Enterprise-grade scalability, security, and seamless integration with Ford's existing technology ecosystem
Advanced natural-language understanding with citation-backed answers
Cross-functional workflow automation spanning engineering, service, and business domains

Technologies Used:

LLMs + semantic vector search • Metadata catalog • Cloud-native micro-services • Secure APIs • CI/CD & MLOps • GPU-accelerated inference

BETA

PRO_TRAILER_HITCH_ASSIST

ADAS feature replacing trial-and-error hitching process with a single-button workflow. Utilizes CNNs and 360-degree sensor fusion.

#ADAS #CV #SENSOR_FUSION

Last updated: 1w ago

> Role: Lead Feature Systems Engineer, ADAS R&A

Pro-Trailer Hitch Assist (PTHA) is Ford’s fully automated hitch-alignment feature, engineered to replace the traditional two-person, trial-and-error process with a single-button workflow. A 180-degree rear HD camera, corner radars, and a 12-sensor ultrasonic array feed a convolutional-neural-network trained on more than 60 000 real-world hitching events, enabling reliable coupler detection in adverse lighting and weather. Once activated, the ADAS controller fuses vision-based pose estimates with wheel-speed and steering-angle data and directs steer-, throttle-, and brake-by-wire actuators through a 25 Hz control loop.

The vehicle follows a decelerating reverse trajectory and stops when ball-to-coupler offset is ≤ 25 mm, then seamlessly hands off to Smart Hitch for tongue-weight validation. Built-in safeguards—including ultrasonic range gating, instant driver override, and self-cancellation above 8 km/h—ensure operational safety, while over-the-air updates continuously refine the machine-learning model from anonymised fleet data. Deployed across F-150, F-150 Lightning, and Super Duty platforms, PTHA has reduced average hitch-up time by more than 80 percent and virtually eliminated the need for ground spotters.

Key Features:

Hands-free hitching – Full drive-by-wire control aligns ball to coupler in seconds
AI vision + radar fusion – Detects hitch components and obstacles up to 20 ft, day or night
Real-time camera overlays – Live guidance graphics keep the driver informed
Smart Hitch integration – Calculates tongue weight and flags load-balance issues
One-touch activation – Start/stop from a single dash button; auto-cancels above ≈5 mph
OTA-updatable – ML model and feature set evolve via Ford Power-Up updates

Technologies Used:

ML vision models • Rear HD camera & corner radars • Ultrasonic array • Sensor-fusion ECU (PAM) • Drive-by-wire steering & brakes • CAN / LIN networking • OTA update pipeline • On-board weight sensing

STABLE

AV_MARSHALLING_PLATFORM

Cloud-orchestrated, Level-4 logistics platform that drives finished autonomous vehicles from assembly lines to yards seamlessly without drivers.

#L4_AUTONOMY #SYSML

Last updated: 1mo ago

> Role: Lead Technology Architect – Autonomous Vehicle Marshalling

Ford’s Autonomous Vehicle Marshalling (AVM) platform replaces end-of-line shuttling crews with a cloud-orchestrated, Level-4 logistics service that drives finished vehicles from the assembly conveyor to test stations, charge bays, and outbound yard slots without human intervention. Infrastructure-mounted LiDAR and ultrawide-band beacons localise each car to centimetre accuracy, while a V2X link streams trajectories from an off-board motion-planning engine that continuously accounts for pedestrians, AGVs, and static obstacles. At the vehicle, existing drive-by-wire actuators execute low-speed path-following under a functional-safety framework validated through ISO 26262 FMEA and fault-injection campaigns.

Tight integration with the plant’s MES/ERP stack lets the dispatcher assign VIN-specific routes, log real-time telemetry, and trigger automatic OTA updates. Pilot deployments at Ford’s Saarlouis and Cologne sites have removed up to 90 % of manual yard moves, reduced cycle time through loading docks by half, and provided live location visibility for every unit on the lot—turning a traditional throughput bottleneck into a fully traceable, autonomous flow of finished vehicles.

Key Features:

Driver-out marshalling – Fully autonomous hand-offs from plant conveyor to shipping yard
Cloud fleet manager – Schedules routes, tracks state of charge/fuel, and optimizes yard flow
Low-speed safety stack – Redundant sensing and ISO 26262-aligned control at <15 mph
Plant-IT integration – MES & ERP APIs sync VIN, build status, and dock assignments
End-to-end traceability – Jama + Syndeia digital thread from requirements to test artifacts
Variant-aware architecture – Product-line engineering supports multi-generation platforms

Technologies Used:

SysML • ROS2 • GNSS / UWB localization • Sensor fusion (LiDAR / Radar / Camera) • Jama & Syndeia digital thread • Jama Connect • Ansys ModelCenter • FMA / FMEA • ISO 26262 & ISO/PAS 21448 compliance

DEPLOYED

DRIVE_BY_WIRE_SYSTEM

Modular suite converting mass-produced vehicles into autonomous-ready platforms. Sub-20 Hz command loop over ROS API.

#DBW #ROS2 #C++

Autonomous Vehicle Platform

> Role: Lead Systems Safety Engineer

Dataspeed’s Drive-By-Wire (DBW) Kit converts mass-produced vehicles into autonomous-ready research platforms by exposing deterministic, software-level control of brake, throttle, steering, and transmission subsystems. A modular suite of by-wire controller modules interfaces directly with the OEM CAN backbone and is joined by a configurable CAN gateway—four independent 1 Mbps networks with breakout for up to five additional nodes - so perception or V2X devices can be added without tapping factory wiring. An intelligent 12-channel power-distribution unit supplies fused, solid-state outputs and supports hard-wired e-stop integration for safety-of-test operations. Commands and feedback run at 20 Hz over a ROS-compatible API, enabling university and industrial teams to drop the kit into their autonomy stacks within hours

Built-in diagnostics, watchdog counters, and driver-override logic ensure the vehicle reverts to OEM authority if thresholds are exceeded, while calibration utilities streamline steering-angle and brake-pressure alignment during installation. Supported on popular development platforms—Ford Fusion / Mondeo, Chrysler Pacifica, Jeep Grand Cherokee, and Polaris GEM—the Dataspeed DBW Kit has become a de-facto standard for AV prototyping, competition teams, and government R&D programmes seeking reliable, repeatable low-level control without the cost of bespoke drive-by-wire retrofits.

Key Features:

Full by-wire control – Brake, throttle, steer, and shift modules with sub-20 Hz command loop.
Multi-CAN gateway – Four independent 1 Mbps networks plus breakout for five add-on nodes.
Intelligent power distro – 12-channel solid-state PDU for safe accessory integration.
Fail-safe architecture – Driver-override detection, ABS/AEB passthrough, and ISO 26262-aligned diagnostics.
ROS & C++ SDK – Publish/subscribe interface lets autonomy algorithms command the vehicle in minutes.
Vehicle-agnostic kits – Supported on popular OEM platforms (Ford Fusion, Jeep Grand Cherokee, etc.)

Technologies Used:

CAN & LIN networking • ROS2 / C++ SDK • Redundant ECU controllers • Solid-state PDU • Functional-safety analysis (FMEA, ISO 26262) • Model-based design (Simulink / SysML) • Hardware-in-the-loop (HIL) validation rigs

STABLE

L4_SELF_DRIVING_PLATFORM

Uber ATG's Level-4 robotaxi platform with 360-degree perception, redundant actuation, and edge-to-cloud ML.

#L4_AUTONOMY #SYSML

Autonomous Vehicles

> Role: Systems Engineer, MBSE

Uber ATG’s Level-4 robotaxi platform is built on specially engineered Volvo XC90 SUVs equipped with a roof-mounted sensor crown that integrates a 64-channel Velodyne LiDAR, seven long- and short-range radars, and fourteen HDR cameras for full 360-degree perception. A dual-computer, fail-operational compute stack in the cargo area fuses these inputs, localises to centimetre-precision HD maps, classifies and predicts surrounding actors, and generates real-time trajectories. Redundant steering, braking, and power domains—each monitored by independent safety controllers—enable the vehicle to maintain control or execute a minimal-risk manoeuvre in the event of a single-point failure, a prerequisite for commercial ride-hailing.

Pilot operations began in Pittsburgh and expanded to San Francisco and Phoenix, collectively logging millions of autonomous miles. Continuous cloud ingestion of petabyte-scale sensor data drives an edge-to-cloud machine-learning pipeline that refines perception and planning models, ensuring the fleet benefits from fleetwide learning while remaining compliant with stringent functional-safety requirements.

Key Features:

Multi-sensor fusion – 360° Velodyne LiDAR, 7 radars, 14 HD cameras feed a real-time perception graph.
Redundant actuation – Dual-path steering & braking architectures halt safely on single-point failure.
HD-map localization – Centimetre-level pose using LiDAR reflectivity and camera semantics
Edge-to-cloud ML loop – Fleet data off-loaded nightly for labeling, training, and OTA model updates.
Scenario-based simulation – Hardware-in-the-loop and ROS-Gazebo rigs replay millions of miles of drives.
Operational design domain – City-scale ODD taxonomy drives structured V&V test matrix.

Technologies Used:

SysML • ROS • HD maps • LiDAR / Radar / Camera stacks • Redundant ECU & drive-by-wire • GPU-accelerated perception • Ansys Medini FMEA / FTA • Jama Connect digital thread • Cloud-scale data labeling & ML training

DEPLOYED

EXTREME_ROBOTICS_PLATFORM

Low-cost open-source robotics ecosystem created by Cornell Cup. RP2350 powered with Python/Blockly SDK.

#RASPBERRY_PI #PYTHON

Educational Robotics

> Role: Lead Systems Engineer

XRP is a low-cost, open-source robotics ecosystem created by Cornell Cup Robotics and an industry consortium that includes FIRST, SparkFun, DEKA, Raspberry Pi, STMicroelectronics, DigiKey, and WPI. The platform scales from K-12 classrooms to graduate research, letting learners assemble a robot in minutes, program it in Blockly or Python, and expand with off-the-shelf Qwiic sensors—all while tapping into a global community of educators and makers.

Key Features:

Tool-free hardware kit – snap-fit chassis, dual-core RP2350 controller, motor drivers, servos, wheels, and casters for rapid build-ups.
Multi-language coding – Blockly for beginners, Python & WPILib for advanced control and FRC® compatibility.
Wireless connectivity – on-board Wi-Fi / Bluetooth via Raspberry Pi RM2 radio for remote driving and IoT projects.
Rich sensor suite – ultrasonic distance, line-tracking array, 6-DoF IMU, and plug-and-play Qwiic ports for color, RFID, or spectral sensors.
Open-source everything – KiCad schematics, Onshape CAD, and permissive licenses encourage customization and community mods.
Learning hub & forum – step-by-step modules, project challenges, and educator discounts tied into the FIRST® network.

Technologies Used:

Raspberry Pi RP2350 MCU • Qwiic I²C bus • Blockly & Python SDK • WPILib • Wi-Fi / Bluetooth connectivity • 6-DoF IMU • Motor encoders • Open-source CAD & KiCad hardware

STABLE

AUTONOMOUS_SURFACE_VEHICLE

Catamaran with CFD-optimized hull and Tri-T200 Thruster Propulsion. Houses an AUV deck launch.

#MATLAB #C_CPP

Marine Swarm System

> Role: Team Lead and Systems Integrator

Trishul is a 1.19 m × 0.85 m fiberglass catamaran whose third-generation hull—shaped through iterative SolidWorks modelling and ANSYS CFX analysis—eliminated 56 % of the drag force by adding complex stream-lining curves while keeping the wetted area nearly constant. Built light (≈ 12.6 kg dry) yet strong, the single-piece lay-up resists seepage far better than the plank-built prototypes that preceded it, and its watertight penetrators let three forward-facing Blue Robotics T200 thrusters vector up to 13 kgf of thrust for sprint speeds to 2.5 m s⁻¹ and precise station-keeping. A recessed deck bay carries “Dolphin,” a four-thruster acrylic AUV, enabling cooperative surface–subsurface missions without altering the vessel’s balance

Below deck, a vertical card-guide backplane hosts seven hot-swap PCBs—power regulation, cooling control, safety interlocks, and more—so failures can be isolated and boards slid out like server blades. Sensor fusion and mission logic run on an Intel i5 SBC with dual-band 5 GHz Wi-Fi, while an ATmega 328P handles auxiliary I/O and a Cypress PSoC-4 delivers 16-bit PWM for thruster finesse. Together, these hydrodynamic, electrical, and software choices turn Trishul into a modular research platform that can dash through speed gates, hold position in currents, and deploy an AUV—all on a hull that skims just 24 cm below the surface when fully loaded.

Key Features:

CFD-Optimized Displacement Hull - complex curvature and reinforced fiberglass lay-up reduce drag while maintaining stability in chop.
Tri-T200 Thruster Propulsion - three BlueRobotics T200s deliver up to 2.5 m s^-1 and fine vectoring for station-keep or high-speed dashes.
AUV Deck Launch - hull integrates a bay to deploy “Dolphin,” a four-thruster acrylic AUV for interoperability missions.
Modular Electronics Stack - seven custom PCBs slide into a vertical-mount backplane for plug-and-play swaps and rapid debug.
Intel i5 Mission Computer - Gigabyte µATX board running C & MATLAB code orchestrates GPS/IMU/LiDAR-fused navigation, PID control, and waypoint autonomy.

Technologies Used:

ANSYS CFX & Static Structural • SolidWorks • Fiberglass lay-up • BlueRobotics T200 thrusters • Intel i5 SBC • ATmega 328P & Cypress PSoC 4 microcontrollers • custom buck/EMI-safe power PCBs • dual-band Wi-Fi telemetry • GPS • IMU • LiDAR • HD camera sensor suite • C / MATLAB modular mission software.

VESSEL.AI COPILOT

Key Features:

Technologies Used:

PRO_TRAILER_HITCH_ASSIST

Key Features:

Technologies Used:

AV_MARSHALLING_PLATFORM

Key Features:

Technologies Used:

DRIVE_BY_WIRE_SYSTEM

Key Features:

Technologies Used:

L4_SELF_DRIVING_PLATFORM

Key Features:

Technologies Used:

EXTREME_ROBOTICS_PLATFORM

Key Features:

Technologies Used:

AUTONOMOUS_SURFACE_VEHICLE

Key Features:

Technologies Used:

PROJECT_TITLE