Voice latency
~1.25s
Reduced from roughly 2-3 seconds to a more natural conversational pace.
Senior Care AI SaaS
Built an AI workforce for caregiver agencies that automates outreach, check-ins, companion calls, and escalations.
Senior Care AI transformed repetitive, high-volume care communication into a configurable service layer across phone, SMS, dashboards, and reports, helping agencies scale outreach without losing empathy, safety, or caregiver visibility.
The core challenge was not just adding AI. It was packaging telephony, scheduling, escalation
rules,
reporting, multilingual communication, and low-latency voice into a product that older adults
could use
comfortably and caregiver teams could trust operationally.
Automation window
24/7
Always-on reminders, inbound support, and escalation-aware workflows.
Engagement length
1.5 yrs
From discovery and architecture through production hardening.
Leadership scope
5-6 devs
Led architecture, execution, delivery practices, and technical direction.
What it solved
- Manual, repetitive calling and follow-up workflows that were hard to scale.
- Missed reminders and weak visibility into risk signals or failed outreach.
- High-value caregiver time spent on low-complexity communication tasks.
What was built
- A multi-tenant senior care automation platform across phone, SMS, dashboards, and reports.
- A low-latency AI companion engine for inbound and outbound voice interactions.
- A configurable workflow and escalation system tuned for agency operations.
Why it mattered
- Care agencies gained 24/7 communication coverage without linear staff growth.
- Elderly users got accessible, voice-first interactions designed around comfort and simplicity.
- Care teams and families received clearer visibility into outcomes, alerts, and follow-ups.
The Challenge
Caregiver agencies needed a way to stay in touch with elderly clients consistently and safely at scale. Typical dashboards did not solve the last-mile communication problem, and manual outreach created gaps in responsiveness, consistency, and operational visibility.
Operational pain points
- Routine calls and reminders were manual, repetitive, and difficult to coordinate reliably.
- Delays in reminders, follow-ups, or escalation handoffs carried real care risk.
- Signals like confusion, distress, missed tasks, or lack of response were easy to overlook.
- Families wanted better visibility without creating more administrative work for staff.
Product requirements
- Support medication reminders, surveys, announcements, and AI companion conversations.
- Stay voice-first and accessible for older adults who may not be app-native.
- Provide configurable escalation logic, reporting, transcripts, and dashboards.
- Operate reliably 24/7 with low-latency voice interactions and clear system observability.
How The System Worked
The product was intentionally designed around operational workflow, not chatbot novelty. Agencies could configure care logic once, then let the platform generate, execute, interpret, and escalate communication automatically.
01. Configure care plans
Define what to send, when to send it, through which channel, in which language, and what counts as a risk signal.
02. Generate scheduled tasks
Create reminders, surveys, announcements, check-ins, follow-ups, and family reports from agency configuration.
03. Execute automatically
Run Twilio voice calls, SMS messages, and AI-led conversations at the right time with no caregiver intervention.
04. Interpret responses
Evaluate outcomes using task type, user profile, prior interactions, and escalation rules instead of raw events alone.
05. Escalate and report
Route missed reminders, help requests, confusion, or distress to caregivers, coordinators, relatives, or multiple contacts.
The Solution
Senior Care AI was structured as a focused set of services spanning caregiver workflows, telephony execution, AI companion orchestration, background job processing, and cost-conscious production deployment. The system balanced flexible conversational AI with deterministic care workflows through task context, prompts, escalation rules, and operational safeguards.
Core product capabilities
- Medication reminders and appointment announcements.
- Automated wellness surveys and recurring check-ins.
- AI companion and inbound support calls for elderly users.
- Family notifications, escalation alerts, dashboards, and reports.
- Multilingual communication with senior-friendly accessibility tuning.
Key design decisions
- Separated user management, telephony, AI streaming, and workers into distinct services.
- Used Celery plus Redis for scheduled jobs, retries, and follow-up workflows.
- Kept PostgreSQL as the source of truth for users, tasks, transcripts, and escalations.
- Used a streaming-first AI model to reduce perceived lag in voice calls.
- Supported multiple LLM providers for quality, cost, and fallback flexibility.
Capability Snapshot
The platform's commercial value came from reflecting the messy realities of care operations rather than offering a one-size-fits-all reminder bot.
Configurable scheduling
Supported one-time and recurring schedules, quiet hours, retries, fallback routes, and family-only notifications.
Accessible communication
English and Spanish support, adjustable communication tone, and voice-first flows suitable for older adults.
Survey intelligence
Branching survey logic, incomplete handling, follow-up rules, and context-aware outcome interpretation.
Companion controls
Configurable frequency, duration, safe topics, and summaries for AI companion conversations.
Operational visibility
Task history, completion state, retry outcomes, transcripts, relative notifications, and escalation reports.
Multi-tenant readiness
Designed for agencies, coordinators, and managers who needed repeatable operations across many elderly users.
Implementation Highlights
These sections are collapsible on purpose so the page stays scan-friendly for a first pass, while still holding enough depth for a client who wants to evaluate the technical and product thinking behind the build.
Care-operations workflow engine This was a real operations product, not a thin chatbot wrapper.
Challenge: Senior care outreach is repetitive, but not simplistic. The product needed to manage schedules, retries, escalation logic, transcript logging, family updates, and reporting in one operational flow.
Approach: The platform converted agency configuration into scheduled tasks, follow-ups, alerts, and reporting workflows, giving care teams one coordinated system for communication operations.
Result: Agencies gained a scalable communication layer that reduced manual admin work while preserving oversight and control.
Telephony plus AI streaming integration Natural phone experiences required more than simply connecting an LLM to Twilio.
Twilio call flows, backend webhooks, context loading, and streamed AI responses had to work together in real time for older adults who expect a calm, understandable phone interaction.
The system combined telephony control with context-aware AI companion behavior so reminder calls, inbound support, and check-ins could feel responsive without losing structure or safety.
Latency optimization for trust In voice care workflows, responsiveness affects comfort and confidence.
Problem: Early voice latency of roughly 2-3 seconds made conversations feel less natural and risked reducing trust in the system.
Solution: A streaming-first interaction model reduced perceived delay and improved the cadence of AI-assisted conversations.
Outcome: Voice latency improved to around 1.25 seconds, materially improving usability and the perceived intelligence of the system.
Configurable escalation logic Different agencies needed different rules, timing, and recipients.
The escalation system was built to be configurable rather than hardcoded. Agencies could define what qualified as a risk event, who should be notified, which channel to use, and when multi-recipient escalation paths should activate.
This allowed the same platform to support diverse operating styles without forking product logic for every client.
Transparent operational visibility Care teams needed evidence, not black-box automation.
The product surfaced completed, pending, and missed tasks; transcript review; escalation severity; relative notification history; and dashboard-level communication trends.
That visibility made the system easier to trust, easier to audit, and easier to improve as agencies expanded adoption.
Results & Business Impact
The product's value came from combining accessibility, operational rigor, low-latency voice, and configurable escalation into one platform that matched the real workflows of senior caregiver agencies.
Agency operations
- Automated repetitive outreach across reminders, surveys, announcements, and check-ins.
- Reduced manual coordination load while improving consistency of outreach execution.
- Enabled agencies to scale service delivery without linearly scaling staff time.
Elderly user experience
- Delivered voice-first interactions designed for accessibility and ease of use.
- Enabled inbound and outbound AI-supported voice workflows with more natural pacing.
- Created safer escalation paths when help requests, confusion, or distress were detected.
Operational trust
- Provided dashboards, reports, transcripts, and history for transparent follow-up.
- Delivered near-real-time escalation handling and family notifications.
- Shipped as a production-oriented multi-service deployment with CI/CD and hardening practices.
My Role & Scope
I worked as the Lead AI Engineer, combining architecture leadership, product thinking, hands-on implementation, and delivery ownership across the full engagement.
Owned directly
- Designed the overall system architecture and service boundaries.
- Gathered and refined requirements with the client.
- Led a team of 5-6 developers across implementation and delivery practices.
- Defined CI/CD, SDLC, production hardening, and reliability direction.
Hands-on execution
- Built backend, AI orchestration, and workflow systems directly.
- Solved performance, reliability, and integration issues as the platform grew.
- Worked across telephony execution, AI streaming, data flows, and deployments.
- Balanced product simplicity for caregivers with backend complexity behind the scenes.
Tech Stack
The implementation favored pragmatic, production-oriented tools that supported real-time communication, background execution, and cost-conscious deployment.
Backend
Python, FastAPI, Twilio Voice and Messaging, PostgreSQL, SQLModel, Alembic, Celery, and Redis.
AI layer
OpenAI, Anthropic, Together AI, prompt orchestration, and Deepgram ASR/TTS for voice interactions.
Deployment
Docker Compose on an InterServer VPS with Nginx, Cloudflare DNS/SSL, GitHub Actions CI/CD, Streamlit, and Flower.
Final Takeaway
Senior Care AI turned manual care communication into a scalable service layer. The value was not just in adding AI, but in combining voice accessibility, operational structure, reporting, and safe escalation paths into a product that matched how real caregiver agencies work.