HemoSave Device

About this project

Postpartum hemorrhage kills thousands of mothers every year, especially in low-resource settings. Existing approaches either rely on rough visual estimation, which can be wildly inaccurate, or on premium devices that are too expensive and slow to deploy at scale.

HemoSave is an ongoing project to close that gap: a device designed to measure blood loss quickly and affordably, targeting a correlation with premium devices while being feasible for hospitals that can't spend thousands per unit. Learn more at bainitech.com.

• Co-founded the project and helped define the core product requirements.
• Ran conversations with clinicians to understand workflow constraints and real failure modes.
• Focused on making the system usable in noisy, resource-constrained environments.

Status: Active • Ongoing validation and iteration.

SantaBot

About this project

SantaBot is a playful robotics project built around a simple idea: can we design a robot that feels delightful but still uses the same core systems we'd rely on for more serious work?

It combines motion, sensing, and behavior in a compact package — a way to practice robotics fundamentals while shipping something that actually makes people smile.

• Implemented behaviors that react to the environment in real time.
• Used this project as a sandbox to test control ideas and hardware integrations.

Status: Complete for now • Future upgrades possible.

Kinematics Library

About this project

A lot of robotics students run into the same wall: the math behind kinematics is already hard, and the libraries they use can feel opaque, under-documented, or too tied to specific hardware. This library is my attempt to fix that — starting from the math, not the framework.

It's built around a SerialChain abstraction that accepts any mix of revolute and prismatic joints, handles symbolic design parameters (link lengths, offsets) separately from runtime joint angles, and exposes both symbolic and high-speed numeric paths.

How it works

A robot is described by listing joints with R() (revolute) or P() (prismatic) calls. Joint origins and axes can be symbolic expressions — so link lengths and offsets stay as parameters until you're ready to fix numbers. The chain pre-compiles home-pose transforms whenever the design changes, keeping per-query FK fast.

# Define a 3-DOF arm with symbolic link lengths
chain = SerialChain(
    joints=[
        R("z", at=(0, 0, 0), q="q1"),
        R("y", at=(L1, 0, 0), q="q2"),
        P("z", at=(L2, 0, 0), q="q3"),
    ],
    ee=("L3", 0, 0),
)

# Numeric FK — single config
pts = chain.fk_points({L1: 0.4, L2: 0.3, L3: 0.2},
                       {q1: 0.5, q2: -0.3, q3: 0.1})

# Symbolic Jacobian — computed once, then cached
J = chain.jacobian_symbolic()

What's built so far

• SerialChain core — forward kinematics (single and batched), SE(3) transforms, home-frame compilation with design-change detection.
• Symbolic layer — SymPy end-effector position expression and geometric Jacobian, useful for deriving closed-form IK or checking singularities.
• Workspace visualizer — Dash app with real-time 3D rendering of joint cylinders, rotation-axis arrows, arc indicators, and a reachable workspace cloud (convex hull if SciPy is available, point cloud otherwise).
• Modular geometry — separate helpers for cylinder meshes, orthonormal frame construction, and arrowhead lines keep the rendering layer decoupled from the math core.

Status: Active • Being expanded as I learn — IK and dynamics coming next.