VOIP Engineer

Hi — thanks for reaching out.

Before we go any further, I want to clearly explain what we’re building right now, how it works, and what “done” looks like.

1) What CallDialer is (layman’s terms)

CallDialer is our high-volume outbound calling machine.

At scale, it does four core things:

Dials millions of phone numbers from lead lists

Uses Answering Machine Detection (AMD) to determine who answers

If a human answers, it plays a message asking them to “Press 1”

If Press-1 is detected, the call is instantly routed/transferred

If a voicemail answers, it can optionally leave a prerecorded message

Think of it as a fully automated system that:

Separates humans from machines

Captures real intent via keypress

Converts outbound calls into qualified inbound demand

2) Critical requirement: Answering Machine Detection (AMD)

AMD is non-negotiable and a core success factor.

CallDialer must be able to:

Detect human answer vs voicemail vs answering service

Decide in real time whether to:

Play the press-1 message

Leave a prerecorded voicemail

Hang up and retry later

Why this matters

Without accurate AMD:

You burn carrier reputation

You leave messages where you shouldn’t

You destroy response rates and CPS efficiency

This is not optional — it must be reliable and tunable.

3) Voicemail delivery verification & intelligence

The system must not blindly assume voicemails are being left successfully.

We expect CallDialer to determine voicemail success using:

CDRs (Call Detail Records)

Call duration thresholds

Audio timing vs voicemail greeting length

Post-call analytics (response rate correlation)

Detection of FAS (False Answering Services)

False Answering Services (FAS)

FAS are systems that:

Answer like a human

Delay speech

Then route to voicemail or IVR

The dialer must be capable of:

Identifying likely FAS patterns

Flagging numbers / routes / carriers with high FAS rates

Adjusting behavior (hang up, retry, exclude, or treat as VM)

This is essential for maintaining true pickup rates and protecting scale.

4) Technical breakdown (how it actually works)

Our repo is already largely built and structured as a production system.

A) API / Orchestrator (Fastify + TypeScript)

The “brain” of the system

Handles:

Campaign execution

Origination logic

CPS throttling

AMD decisioning

DTMF event handling

Transfer triggers

Logging & metrics

Key areas:

apps/api/src/routes/*

apps/api/src/services/*

apps/api/src/telephony/* (ESL client, call state, AMD/DTMF logic)

B) Telephony Engine (FreeSWITCH)

Executes real outbound calls via SIP gateways

Controlled via ESL (Event Socket Library)

Emits real-time events:

Answer

Hangup

DTMF digits

Media state

Call duration

This is where:

AMD signals originate

DTMF digits are captured

Transfers are executed (uuid_transfer, bridge logic, etc.)

C) Optional SIP Proxy (Kamailio)

Present for:

NAT handling

SIP normalization

Future carrier expansion

High-scale routing flexibility

D) Data Layer (Postgres + Redis)

Postgres: campaigns, leads, attempts, CDRs, outcomes

Redis: rate limiting, queues, state coordination at scale

5) Press-1 / DTMF ? Transfer flow (core revenue path)

This must be flawless.

Flow:

Human answers

Message plays

User presses 1

DTMF detected instantly

Call is transferred immediately

All events are logged correctly

Latency, missed digits, or dropped transfers are unacceptable.

6) Carrier connectivity expectations

CallDialer must connect directly to carriers using:

SIP (primary / expected)

Carrier APIs only if absolutely required

This is not Twilio or CPaaS abstraction.
This is real carrier-level dialing.

7) Execution sequence (important)
Phase 1 — Immediate

One carrier

~100 CPS

Max it out

Stabilize AMD, DTMF, voicemail, transfers

Prove reliability

Phase 2 — LCRHV

Only after Phase 1 is stable

Multi-carrier routing for cost optimization

LCRHV is deferred unless required to stabilize Phase 1

Datamanager & CallHub

Part of the ecosystem

Not blockers for CallDialer completion

8) Performance definition (“done” means this)

2,000,000 dials/day sustained

Accurate AMD

Verified voicemail delivery

FAS detection and mitigation

Clean CDRs and outcome attribution

Reliable real-time press-1 transfers

Production-grade monitoring, retries, and failure handling

9) Required development workflow (non-negotiable)

Work must be done inside Cursor or Antigravity

Manual-only coding workflows are not acceptable

Speed, iteration, and testing velocity matter

10) What to include in your reply

If aligned, reply with:

Experience with FreeSWITCH, ESL, SIP gateways, or large-scale VoIP

Experience with AMD, voicemail detection, or FAS issues

Experience handling DTMF ? transfer flows

Confirmation you can work fully inside Cursor or Antigravity

Any immediate concerns about scaling to 2M dials/day (and how you’d address them)

If this aligns, we’ll move quickly.

Thanks.