Haseeb ArshadHaseeb Arshad

Integration surface

APIs I build against

Most of the difficulty in a conversational AI product is not the model. It is everything the model has to reach: a PBX that emits events in its own shape, an EHR that owns the only real appointment slot, a CRM that must not end up with duplicate records, a payment gateway where a retry costs money. These are the platforms I have integrated, and what each one is actually used for.

Platforms
64
Categories
8
Years shipping
7+
Auth patterns
OAuth 2.0

Voice and telephony

12 platforms

Carrier, PBX, and meeting platforms behind live call handling: inbound routing, call events, recordings, transfers, and fallback paths.

  • SIP and SBCPolyAI

    Call paths, routing behaviour, transfers, and fallback destinations validated per environment.

  • GoTo ConnectPeerLogic

    Call and voicemail events consumed through webhooks into conversation pipelines.

  • NetSapiensPeerLogic

    Hosted PBX events and call metadata normalized behind a single adapter.

  • RingCentralPeerLogic

    Telephony events and messaging APIs wired into practice workflows.

  • BandwidthPeerLogic

    Programmable voice and SMS for patient messaging and call control.

  • AsteriskPolyAI

    Open-source PBX used in SIP call paths and migrations.

  • KamailioPolyAI

    SIP proxy and routing layer in front of production voice traffic.

  • Programmable messaging and voice primitives for conversational flows.

  • Voice and messaging APIs for carrier-side call handling.

  • Practice communications platform data joined to scheduling workflows.

  • ZoomPolyAI

    Meeting and phone APIs integrated into assistant workflows.

  • Enterprise calling and collaboration endpoints for internal routing.

Healthcare and EHR

9 platforms

Practice-management and records systems: real appointment availability, patient context, and write-backs that have to be right the first time.

  • NexHealthPeerLogic

    Appointment availability, booking, and patient sync behind scheduling agents.

  • Open DentalPeerLogic

    Practice-management data read and written through its API for dental workflows.

  • Specialty EHR records surfaced to conversational workflows.

  • Ambulatory EHR data and scheduling endpoints.

  • Pharmacy management data for prescription-aware conversations.

  • Veterinary practice software integrated for appointment workflows.

  • Integration middleware used to reach practice systems without one-off connectors.

  • Online scheduling data joined to voice and SMS booking.

  • FHIRPolyAI

    Standards-based resource modelling for records, appointments, and coverage.

AI and model APIs

9 platforms

The model layer: hosted and local inference, agent frameworks, speech, and the evaluation harnesses that gate a release.

  • Frontier model calls for drafting, extraction, and reasoning, schema-bound and routed per task.

  • OllamaHomelab

    Local model serving on a GPU workstation for private, zero-marginal-cost inference.

  • LangGraphPeerLogic

    Stateful multi-agent graphs for scheduling, follow-up, and escalation.

  • LangChainPeerLogic

    Retrieval and tool-calling plumbing around production LLM workflows.

  • Tool calling and RAGPolyAI

    Python functions and retrieval scoped per caller so answers come from facts, not pattern matching.

  • Speech synthesis and custom brand voices for production agents.

  • Recognition models evaluated and tuned for programme-specific vocabulary and accents.

  • Tools and telemetry exposed to agentic workflows over a standard protocol.

  • Local speech-to-text for transcription and evaluation fixtures.

CRM, hospitality, and business systems

8 platforms

Systems of record that a conversation has to read from and write back to without creating duplicate or orphaned records.

  • Records, cases, and activity written back from live assistant workflows.

  • Contact and deal objects synced with outreach workflows.

  • Ticket creation and status lookups for support handoff.

  • Conversation and contact APIs for messaging handoff.

  • OpenTablePolyAI

    Reservation availability and booking actions called from live conversations.

  • ZonalPolyAI

    Hospitality POS and booking data joined to voice workflows.

  • Event and group booking records created from assistant actions.

  • Enterprise request and incident records created from assistant actions.

Google and messaging APIs

8 platforms

Mail, calendar, storage, and chat surfaces that carry work in and notifications out, all on OAuth 2.0 with least-privilege scopes.

  • Gmail APIHomelab

    Inbox triage, threaded replies, and approval-gated sending with attachments.

  • Availability lookups and booking behind a self-hosted scheduling front end.

  • Document import into a private corpus, with change detection and dedupe.

  • Consent, refresh, and token storage handled so integrations survive a week without a human.

  • Operational digests and approval prompts pushed to a phone.

  • IMAP and SMTPHomelab

    Self-hosted mail paths for services that must not depend on a vendor inbox.

  • Webhooks and bot messages for delivery and alert routing.

  • Transactional mail delivery with bounce and event handling.

Cloud and platform

8 platforms

Where the integrations run: managed cloud, containers, and the data stores that keep conversation state consistent.

  • AWSPolyAI

    Lambda, API Gateway, S3, RDS, and IAM behind production services.

  • Google CloudPeerLogic

    Cloud Functions, Pub/Sub, and storage for event-driven integration work.

  • Enterprise-hosted services reached from customer-side deployments.

  • Workload API used directly for rollouts, health gating, and operators.

  • Image builds and registry APIs in the delivery path.

  • Provider APIs driven as code for reproducible infrastructure.

  • Primary store for conversation, scheduling, and application state.

  • Queues, locks, and short-lived state between call turns.

Delivery and observability

7 platforms

The APIs that ship the work and tell me when it breaks: build pipelines, GitOps reconciliation, metrics, and alerting.

  • Build, release, and digest-promotion workflows driven through the API.

  • Argo CDHomelab

    Declarative reconciliation and sync status queried as part of every deploy.

  • PrometheusHomelab

    Custom application metrics exposed and queried for alerts and dashboards.

  • GrafanaHomelab

    Dashboards provisioned as code alongside the services they watch.

  • DatadogPolyAI

    Traces, monitors, and conversation-level telemetry for live AI agents.

  • Error capture and release tracking across services.

  • Issue and delivery workflow automation.

Payments and billing

3 platforms

Money movement, where a retry is never free and idempotency is the whole design.

  • Checkout, subscriptions, and webhook reconciliation.

  • PaystarDatics AI

    Government and utility payment processing in a tax administration platform.

  • NMIDatics AI

    Gateway integration for card processing and settlement reporting.

Where the work happened

PolyAI

Enterprise voice agents

PeerLogic

Patient engagement platform

Datics AI

Client delivery

Homelab

Private AI and platform tooling

Every integration above sits behind an adapter with explicit timeouts, retries, and failure behaviour, so a slow or broken dependency produces a useful outcome for the person on the call instead of silence or an invented answer. Details are in thePolyAIandPeerLogiccase studies.