Field Operations Engine: Technical Architecture & AI Integration Scope

A briefing document outlining Kivun Systems' production dispatch engine, engineering topology, and target LLM integration workflows for consulting partners.

Omer Yaakobi

Head of Engineering • Kivun Systems Ltd.

Confidential Technical Brief

Notice: Prepared for authorized external AI consultants and educators to structure hands-on coding exercises for our 11-developer team.

1. Engineering Team Topology

Kivun Systems employs 38 people. Our engineering department consists of 11 full-time developers divided across three agile squads:

Core Dispatch Engine (4 Engineers): Owns the central constraint solver, PostgreSQL/PostGIS spatial indexes, state-machine transitions, and live ETA calculations.
Integrations & Ingestion (4 Engineers): Manages high-throughput WhatsApp Cloud API webhooks, SMS gateways (Twilio/Infobip), and external ERP connectors.
Dispatcher Console & Mobile Web (3 Engineers): Builds and maintains the React real-time dispatcher console, WebSocket telemetry streaming, and field technician mobile web app.

2. Production Technology Stack

Our stack is modern, containerized, and event-driven, built for sub-second telemetry:

Domain	Core Technologies	Production Role
Backend Services	`Python 3.12`, `FastAPI`, `Asyncpg`, `Node.js`	REST APIs, WebSocket managers, channel webhook processors.
Persistence & Cache	`PostgreSQL 16` + `PostGIS`, `Redis 7 Cluster`	Spatial fleet indexing, ACID state transactions, live session pub/sub.
Messaging & Queues	`RabbitMQ`, `Celery`	Asynchronous task dispatch, retry queues, external webhook fan-out.
Frontend / Mobile	`React 18`, `TypeScript`, `TailwindCSS`	Real-time dispatcher schedule grid, live map rendering.

3. Target Workshop Learning Objectives

We are seeking practical training to empower our developers to build and maintain intelligent agents within our existing architecture:

1 Unstructured Triage & Metadata Extraction

Parsing incoming customer WhatsApp/SMS messages into strongly-typed Pydantic schemas (urgency, skill tags, geo-location, equipment model).

2 Deterministic Tool Calling & DB Interfacing

Implementing structured agent tool calling against our PostgreSQL schema with parameter validation, permission boundary enforcement, and rollback handling.

3 Agentic Exception Handling in Dispatch Loops

Using LLM reasoning agents as fallback supervisors when algorithmic solvers encounter edge cases (e.g. technician delayed, required part missing from inventory).

4 Evaluation, Guardrails & Token Latency

Benchmarking prompt regression, response latency, and preventing hallucinations in SLA-critical field assignments.

Technical Coordinator

For curriculum alignment, NDA execution, or pre-workshop code access:

Omer Yaakobi • Head of Engineering

omer.yaakobi@kivun.site • https://kivun.site