SR Logo/AI Engineering
Generative AIONGOING

ASTRION: AI Logistics Support Agent

Source-grounded support agent for logistics operations that answers order, ticket and policy questions from SOPs and signed customer agreements, computes fees and credits in code, and changes nothing without confirmation.

CategoryGenerative AI
Statusongoing System
Core Tooling
RAGAgentic Tool UseSource CitationsHuman-in-the-LoopNext.jsVercel

01. Problem & Operational Context

Logistics support answers depend on order records, open tickets and contract terms, and figures such as cancellation fees and service credits must be exact. A model that writes those numbers itself, cites nothing, or changes records unprompted cannot be trusted or audited.

02. Technical Approach

Built a support agent that grounds every answer in retrieved policies, SOPs, product documentation and signed customer agreements with each source shown, looks up orders, tickets and accounts at request time, calculates cancellation fees and service credits in code rather than in the model, and requires explicit confirmation before anything is changed.

03. System Architecture & Pipeline

End-to-End System Pipeline & Computation Flow

A workspace-scoped agent that combines cited document retrieval with request-time record lookups, keeps monetary calculations in deterministic code, and gates every state change behind user confirmation.

STEP 01Workspace scoping / ASTRION API
Workspace-Scoped Context

Each session runs as a chosen actor against the accounts attached to the workspace on the server, so lookups only reach records that are in scope.

STEP 02RAG / Source citations
Cited Policy Retrieval

Answers are drawn from policies, SOPs, product documentation and signed customer agreements, with every source shown alongside the response.

STEP 03Agentic tool use
Request-Time Record Lookup

The agent queries orders, tickets and accounts when a question is asked rather than relying on pre-loaded answers.

STEP 04Code-computed / Human-in-the-loop
Deterministic Calculation & Confirmed Actions

Cancellation fees and service credits are computed in code, and no change is made without explicit user confirmation.

04. Key Implementation Details

  • Grounded responses in policies, SOPs, product documentation and signed customer agreements, showing the source behind every answer.
  • Resolved orders, tickets and accounts at request time through tool lookups against the ASTRION API instead of answering from static text.
  • Moved cancellation-fee and service-credit arithmetic out of the model and into code so the figures are calculated, not generated.
  • Required explicit confirmation before any record is changed, keeping the operator in control of state changes.

05. Architectural Tradeoffs & Decisions

Decision: Calculate cancellation fees and service credits in code instead of asking the model to write them.
Justification

Monetary outcomes stay deterministic and reproducible, and the model is limited to explaining the result with cited sources.

Alternative Considered

Letting the language model compute and state the figures directly from retrieved policy text.

07. Evaluation & Observations

Evaluation Methodology:

Exercised through scenario prompts covering fee-free order cancellation, failed-pickup service-credit eligibility, carrier-fault late pickups, stale booking status after collection, and ticket investigation with escalation.

Key Empirical Finding:

Separating retrieval, record lookup, calculation and confirmed action keeps each answer traceable to a source or a computed value.

Verified Artifacts
architectureGrounded Support Agent Flow

Workspace scoping, cited document retrieval, request-time record lookup, code-computed fees and confirmation-gated changes.