SaaS for waste hauling companies, where I've built and maintained multiple
critical features, helping businesses service their clients.
● Route Management
Used by dispatchers to create daily service routes for client orders used by
their drivers. Including route optimization APIs, recurring services, auto
dispatch and comprehensive route and stop management with draft support.
● AI Communications Center
Used by the business's employees to communicate with their clients and drivers
using SMS and phone calls. Handling thousands of calls weekly.
- Using Twilio I built the backend for a complete call center solution featuring
queue management, call recording,
live incall transcriptions, call routing to different departments and agents,
call transfers,
IVR module for creating custom menus callers go through, voicemail.
- Integrating Twilio's client SDK I built the frontend for answering and making
calls.
Implemented a custom websockets solution to synchronize realtime updates to the
UI for calls and messages.
- Using OpenAI's API I built a custom AI agent for
taking taking calls, answering webchat inquiries and replying to SMS messages,
integrating it with the existing system. Customizable by tenants.
- Testing was a challenge since regular prompt eval platforms only test a single
prompt to a single response, but I needed to test how the agent would perform in
arbitrary conversations. Implemented a test suite with custom tooling for
running conversation scenarions and evals on prompt
versions using custom scorers, and a custom prompt optimizer using those
scores.
- Including other features like being able to answer a call while not even on
the page,
headset integration for allowing the use of call center type headset devices to
answer calls.
- Being a critical feature for businesses to reliably communicate with their
clients and not have a single dropped call or missed SMS, I designed the system
with observability in mind, featuring custom logging per call, internal
analytics dashboards, as well as complete unit tests and end to end real world
tests of the entire system.
● Google Meet / Slack clone
Built a Slack/Google Meet clone for internal team communication for the
business's employees using LiveKit.
- Featuring realtime chat in channels/DMs, threads, forwarding, reactions,
notifications.
- Group and one on one huddles with video, video effects, screen sharing,
speaking
indicators, reactions, hand raising, in huddle chat for meeting notes and PiP.
- Optimized the implementation to reduce API usage saving thousands in monthly
costs by utilizing Socket.io.
● Quickbooks Online & Desktop Integration
Inherited two distinct accounting modules implemented in QBO and QBD.
- I had support each tenant's accounting by extensively troubleshooting and
resolving sync issues between the two platforms, often times finding the root
cause and fixing it.
- Dramatically reduced paid API usage by implementing a solution to mirror and
continously sync Quickbooks data to use instead of the API, including
observability dashboards for monitoring usage and service status for each
tenant.
Show More
Stack:
TypeScript, Express, PostgreSQL, Inngest, Twilio,
LiveKit, OpenAI
React, Tanstack
