AI Agents

AI Agents That Work Alongside Your Hospital Team

Most 'AI in healthcare' is a chatbot bolted onto a website. We built agents that analyze data overnight and prepare the morning brief, help your staff work 3x faster, and talk to patients on WhatsApp in their language. Three modes, one platform.

0
Agent Modes -- autonomous, assisted, conversational
0+
Clinical Protocols grounding every response
Multilingual
Hindi, English, and extensible to any language
Any LLM
Pluggable backend -- swap models without changing agents
Auto
Tools auto-generated from schemas -- no manual wiring
24/7
Always-on patient engagement via WhatsApp and Voice

Three Agent Modes

Different jobs, different agents. Same AI Platform, same tools, same safety guarantees underneath.

AU

Autonomous Agents

They work while you sleep

Backend supervisor agents that run without human interaction. Cron-driven and event-driven. Care gap analysis, action prioritization, data quality monitoring -- decisions made overnight, ready for your morning brief.

  • Nightly care gap scans across entire patient population
  • Prioritize and schedule follow-up actions automatically
  • Data quality monitoring and anomaly detection
  • Distributed locking and retry with backoff
  • Cooldown periods to prevent patient message fatigue
AS

Assisted Agents

Your staff, but 3x faster

CRM-embedded specialists that work alongside hospital staff. Reception, Doctor Lookup, Booking, Branch, Emergency -- each with dedicated tools. The agent assists; the human decides.

  • Specialist agents with coordinator routing
  • Generative UI: tool results rendered as React components
  • Live edit mode on active patient records
  • Multi-agent orchestration with context handoff
  • Knowledge bank for live call support
CV

Conversational Agents

Your patients talk to them directly

Patient-facing AI on WhatsApp and Voice. Appointment booking, lab results, medication refills. Safety guardrails enforced at every interaction -- these agents never diagnose.

  • WhatsApp: 24-hour window compliance, session persistence
  • Voice AI: real-time conversations in Hindi and English
  • Appointment booking and care gap follow-up
  • Never diagnoses, always escalates emergencies
  • Complete audit trail of every patient interaction

CRM AI: Multi-Agent Orchestration

One coordinator, multiple specialists. Tools return React components, not just text. Agents operate on live records -- no copy-paste from a chat window.

MO

Multi-Agent Orchestration

One coordinator agent receives all queries and routes to the right specialist. Each specialist has dedicated tools and system prompts. The patient doesn't know they are talking to five agents.

GU

Generative UI Tools

Tools return React components, not just text. Appointment cards, patient summaries, care gap visualizations -- rendered live from tool results. The AI doesn't just answer; it shows.

AC

Agent CMS

WYSIWYG agent builder. System prompt, tool selection, persona, behavior rules. Build a new specialist agent without writing code. We use it internally too.

LE

Live Edit Mode

The agent operates directly on active records. No confirmation dialogs. It reads the record, identifies the action, executes. This is what makes it feel like a coworker, not a chatbot.

TS

Tool Testing Sandbox

Test any tool in isolation -- validate inputs, check outputs, debug edge cases. Every tool is independently testable before you wire it into an agent. We learned this the hard way.

SK

Skills Library

Diabetes care protocols, hypertension management, cardiac risk scoring. Agents load relevant clinical skills on-demand. Every response grounded in published clinical guidelines.

WhatsApp AI: Patient Self-Service

Patients book appointments, check lab results, and manage medications on WhatsApp. 24/7, in their language. Your front desk didn't have to be involved.

Capabilities

Contact Resolution
Identify patient from phone number, match to golden record, load full context
Session Persistence
Maintain conversation context across messages within the 24-hour window
Appointment Booking
Check availability, book appointments, send confirmations -- all via chat
Lab Result Queries
Patients ask about their latest test results and get clear, safe summaries
Medication Refills
Request prescription refills and get routed to the right department
Multilingual
Automatic language detection and switching -- Hindi and English in production, extensible to any language

Safety Guardrails

Patient-facing AI without guardrails is a lawsuit waiting to happen. Every interaction is bounded, monitored, and auditable.

Never Diagnoses
The agent never provides diagnoses or treatment recommendations. Health queries are redirected to clinical channels. No exceptions.
Emergency Escalation
Emergency keywords and sentiment indicators trigger immediate human escalation. Full conversation context transferred. No delay.
24-Hour Compliance
WhatsApp Business API window compliance. Template messages for re-engagement. Automatic session expiry management.
Audit Trail
Every message, tool call, and action is logged. Complete traceability from patient query to system response.

Backend Supervisor: Autonomous Execution

These agents analyze care gaps, prioritize follow-up actions, and flag data quality issues. No human in the loop. That is the point.

Execution Modes

Cron-Driven
Nightly care gap analysis, weekly quality reports, monthly population analytics. Runs on schedule. Decisions ready by morning.
Event-Driven
New patient registered? Lab result arrived? Appointment canceled? The supervisor reacts in real time.
Interval-Based
Continuous monitoring with cooldown periods. Watches data streams, detects anomalies, fires alerts. Like a night shift that never gets tired.

Reliability Guarantees

Distributed Locking
One instance per supervisor, always. No duplicate executions even across multiple servers. We have been bitten by this before.
Cooldown Periods
Configurable cooldown between patient contacts. A patient who gets three messages in one day will opt out. Cooldowns prevent that.
Retry with Backoff
Failed operations retry with exponential backoff. Persistent failures go to a dead letter queue for manual review.
Action Logging
Every autonomous action logged: what triggered it, what was decided, what executed, what resulted. Full traceability.

Safety at Every Layer -- Not Just the Prompt

TB

Topic Boundaries

Scope enforcement

Each agent has a defined scope. Ask it something outside that scope and it redirects or escalates. No creative improvisation.

RedirectEscalate
ET

Escalation Triggers

Hardcoded safety

Emergency keywords, sentiment spikes, clinical risk indicators -- all trigger immediate human escalation. Not configurable. Hardcoded.

KeywordsSentimentRisk
OV

Output Validation

Schema-enforced

Every LLM output validated against strict schemas before execution. Malformed or out-of-bounds responses get rejected and retried. The system does not trust the model blindly.

Schema CheckRetry
AL

Audit Trails

Complete record

What was asked, what was proposed, what was approved, what executed, what resulted. Complete record. Every time.

LoggedTraceable

How THB AI Prevents Hallucination

Typical Healthcare AI (Dangerous)

  • ×LLM generates SQL directly from natural language
  • ×No validation layer between LLM output and database
  • ×Results look plausible but may be clinically wrong
  • ×Hallucinated statistics presented as medical facts
  • ×No audit trail for AI-generated recommendations

THB AI (Grounded)

  • LLM reasons over pre-computed, verified metrics
  • Strict schema validation on every state mutation
  • Planning docs provide clinical protocol grounding
  • Tool outputs are typed and testable independently
  • Complete audit trail: prompt, tools, actions, results

See AI Agents in Action

We will show you all three modes -- autonomous care gap closure, CRM-embedded assistants, and patient-facing WhatsApp AI. Real data, real safety, real outcomes.