Observability & fraud detection
Real-time CDR/SIP/media monitoring, nine detection engines, multi-brand ticketing, and a carrier CRM with SLA enforcement.
Production AI agents that triage fraud, optimise routing, handle customer operations, and answer inbound calls — plus the carrier-grade voice and messaging infrastructure they run on. Built on the best of open source, with high-performance surfaces in Go, Rust, Python, and TypeScript and the latest frontier models on top.
A small team with deep ownership of every layer — from the switches to the agents that operate them. Thirty-five production repositories, thirteen distinct disciplines, one inbox.
Nine purpose-built detection engines watch every CDR, SIP flow, and media stream a carrier produces — surfacing the anomalies that cost margin, quality, and trust. AI agents triage every detection, draft the carrier email, and execute remediations through the same MCP tool surface — under human approval for destructive actions. Detections feed multi-brand ticketing with SLA enforcement and a vendor-and-client CRM with contract terms and maintenance windows.
Each card below is a body of production code we own, operate, or license to carriers. Names of products and customers are intentionally absent — what matters is what each one does.
Real-time CDR/SIP/media monitoring, nine detection engines, multi-brand ticketing, and a carrier CRM with SLA enforcement.
SIP termination, quality-weighted LCR, STIR/SHAKEN, redundant edges with sub-second failover.
Self-service control planes for distributors and resellers; over 500 management tools; row-level tenant isolation.
Replacement consoles for aging contact-center back ends. Built-in MCP, in-app copilot, CRM adapters, WebRTC softphone.
Robo, press-1, power, predictive, preview, progressive — with FTC/TCPA pacing and abandonment controls.
Wholesale aggregation over SMPP and HTTP, rule-driven routing, templates, SMS packs, DLR/MO/EDR pipelines.
Ordering, porting (LOA/FOC), E911, CNAM, STIR registration, supplier reconciliation, and downstream invoicing.
MCP servers exposing 600+ operator tools across Class 4 and Class 5 ecosystems — role-gated, audit-trailed.
Autonomous agents handling NOC triage, route optimisation, customer ops, and inbound voice. Built on the latest frontier models with MCP-mediated tool access, vector memory, and per-tenant isolation.
Branded SIP softphones for contact-center agents — Windows desktop on PJSIP plus a WebRTC client embedded in operator UIs.
Kamailio + RTPEngine with TLS/SRTP, dispatcher failover, and HEP-based observability across clusters.
LERG, DNC, DNO, and FTC complaint datasets refreshed daily and served behind a token-secured download API.
MCP-orchestrated test calls, CDR polling across edge nodes, and automated quality reports for every interconnect.
We compose, we don’t reinvent. Proven open-source cores wrapped in high-performance modern surfaces — Go and Rust where latency matters, Python and TypeScript where iteration matters, and the latest frontier models where intelligence matters.
Carrier-grade work demands carrier-grade discipline. These have been the same since day one and they are not going to change.
We design for the night shift. Health endpoints, idempotent tasks, structured logs, replayable queues, and quiet failovers.
We treat call-detail records, rate decks, and billing reconciliation as primary artifacts. Margins live or die in those files.
The modern parts wrap stable cores. Postgres, Kamailio, FreeSWITCH, Asterisk, RTPEngine, PJSIP, Django, Next.js, Go.
One inbox, no forms, no chat widgets, no calendar links. First reply within 4 hours during 09:00–22:00 UTC.