SS
ShockSmart
Anesthesia-aware ECT decision support
ECT optimization, made visible

Clarity for clinicians.
Progress for patients.

ShockSmart unifies anesthetic choices, seizure quality signals, and key clinical history into one calm, readable view — so the team can focus on the patient instead of juggling spreadsheets and scattered notes.

Real anesthetic regimen based on expert opinion
Users can see the decision trees, our expert rules, and the field test results all in one place.
Designed for high-acuity depression and catatonia, concurrent medical illness, and severe mental illness

ShockSmart is an investigational decision-support prototype — intended to complement, not replace, clinical judgment or institutional protocols.

How well our product works
Macro F1
0.46
▲ How well we perform all classes equally
Weighted F1
0.68
▲ How well we perform overall, giving more importance to the classes that appear more often
Hamming loss
↓ over time
▲ How often our predictions are wrong.
Rare combinations
Low-volume regimens are flagged for manual review and future training, instead of disappearing into the aggregate.
Internal benchmark only · not yet a validated clinical device Next step: Train this model on real-world cases →
ShockSmart anesthetics decision tree dashboard screenshot
◎
Single-glance clarity
Seizure indices, anesthetic choices, and key vitals in one frame, so the team can prepare, deliver, and document without hopping across screens and systems.
Less friction · more shared context
AI
Metrics that surface limits
Macro and weighted F1 are tracked as the model sees more cases, highlighting both where performance is strong and where it is still blind — especially around rare patterns.
Benchmarked, not black-box
↻
Iteration as a feature
Early versions intentionally treat rare or one-off regimens as “learning opportunities,” inviting expert review instead of hiding edge cases behind a single summary score.
Stepwise training · better over time
Model metrics

What the model knows — and what it doesn’t yet.

ShockSmart is in an internal benchmarking phase. Right now, the model is tuned toward common anesthetic combinations, with explicit visibility into gaps around rarer regimens.

20% holdout set Multi-label performance Internal R&D benchmark

On a 20% holdout set, macro F1 sits at 0.46 and weighted F1 at 0.68 — reflecting stronger performance on patterns your team uses most often, with honest drop-off in the tails.

Rare or single-occurrence treatments currently show near-zero precision and recall. Instead of smoothing that away, ShockSmart exposes these blind spots so they can be prioritized in future training cycles.

These numbers are guideposts, not clinical validation. As real-world data, feature engineering, and expert feedback grow, the intent is to re-evaluate performance prospectively rather than freeze the model in its early state.

Macro F1 · 0.46 A sense of how fairly each label is handled, including less common agent combinations.
Weighted F1 · 0.68 Emphasizes performance on the anesthetic patterns most frequently encountered in practice.
Hamming loss ↓ Fewer mismatches between recommended and optimal label sets as training progresses.

Early work has leaned on standardized data due to funding limits. The next step is partnering with ECT programs to pair these metrics with carefully collected, real-world outcomes.

Workflow

A dashboard that fits beside your process, not over it.

ShockSmart is meant to sit next to your existing notes, anesthesia record, and EEG — a light layer that clarifies decisions without asking you to relearn how to practice.

  • Step 01
    Start with the case in front of you
    Enter core clinical factors and a proposed anesthetic plan. ShockSmart mirrors how your team already thinks through risk, rather than imposing a new checklist.
  • Step 02
    See projected context before you commit
    Review projected seizure quality context, hemodynamic considerations, and where the model is more versus less confident, as one more perspective at the table.
  • Step 03
    Close the loop with outcomes
    After the procedure, outcome data can be documented as usual — and, in pilot collaborations, selectively fed back into ShockSmart so edge cases become the next training set.

Quiet UI. Intentional reflection.

The aesthetic is deliberately restrained: soft metallic blues, clean type, and a layout that keeps the patient story, not the software, at the center.

Where ShockSmart adds friction is in reflection — prompting teams to flag outlier cases, annotate rare combinations, and decide together how much weight to give the model’s voice in each decision.

Curious whether ShockSmart fits your ECT program?
We’re partnering with a small number of centers to shape ShockSmart with real-world data and expert consensus. If you’d like to explore a pilot or a co-designed prospective trial, we’d love to talk.
No spam, no auto-enrollment — just a short discussion about your current ECT workflow and whether ShockSmart could add value without adding noise.