Clinical AI — Built from Practice

The chart has
the answer.
We find it.

TrenchesAI builds AI tools for perioperative medicine — starting with a conversational chart navigator that learns how clinicians work. Physician-designed. Built for the clinical reality of complex patients and incomplete charts.

The problem

Preoperative assessment
is broken at scale.

Every surgical patient needs a complete preoperative evaluation. The chart contains the information. But finding it, synthesizing it, and flagging what's missing takes 25–30 minutes per case — and that time compounds across thousands of surgeries every month.

01
Chart review is manual and slow

Preoperative clinicians navigate multiple Epic tabs, unreliable problem lists, disappearing e-consults, and outside records that don't integrate cleanly — every case, every time.

02
Critical findings get missed

Abnormal studies with no follow-up. Specialists seen but no perioperative guidance documented. Medications stopped when they shouldn't be. These omissions create risk.

03
Documentation quality varies

Without a consistent framework, note completeness depends on who's reviewing, how much time they have, and how complex the chart is. Standardization is elusive.

04
Capacity is the ceiling

As surgical volume grows, preoperative medicine is the bottleneck. The answer isn't more clinicians — it's giving the clinicians you have better tools.

The data is already there.

Peer-reviewed benchmarking studies show AI with retrieval-augmented generation outperforms human clinicians on structured perioperative reasoning tasks. The question is not whether this works — it's whether it's built correctly.

96%

Accuracy on surgical fitness assessment vs. 86% for human anesthesiologists

Ke et al. — npj Digital Medicine, 2025

~6min

Time saved per case with AI-assisted preoperative documentation

Ke et al. — npj Digital Medicine, 2025

$146K

Projected annual savings per institution from AI-assisted perioperative workflows

Ke et al. — npj Digital Medicine, 2025

Note generator first.
Active retrieval
second.

When a clinician opens a POM visit, PRISM reads the chart and surfaces a structured draft note — relevant information pulled, gaps flagged, action items surfaced. The clinician reviews, edits, and signs. In Phase 2, the clinician interacts directly with PRISM and the chart in natural language while reviewing the draft — querying, retrieving, verifying. Every interaction is logged, and the system learns over time.

  • Automated note generation

    PRISM reads the chart and produces a structured preoperative assessment draft the moment a POM visit opens. Relevant history surfaced, gaps flagged, action items listed. Clinician reviews, edits, signs.

  • Human-governed protocol floor

    A defined set of clinical standards every note must address. The system flags gaps. The clinician decides. The floor is the same for everyone — the path to it is theirs.

  • Active retrieval layer

    While reviewing the draft, clinicians interact with PRISM in natural language — pulling echoes, cardiology notes, prior anesthesia records, clearance documentation. The navigator emerges from the review workflow.

  • Perioperative continuity

    The same platform follows the patient from POM to the OR to PACU. Every clinician who touches the case uses the same system. Every interaction logged. The system learns and improves over time.

Contact

Built for health systems
ready to move.

We're working with perioperative medicine programs to validate this approach. If you're a health system, clinical leader, or engineer interested in what we're building, we'd like to hear from you.

contact@trenchesmd.ai