Custom AI Agents
AI Customer Support Agent
A retrieval-grounded support agent handling FAQs, order status and product questions, with escalation to the human team.
- Client
- E-commerce retailer
- Year
- 2026
- Services
- Custom AI Agents, Prompt Engineering
70% fewerSupport tickets
24/7Availability
MultilingualLanguages
Figures as reported by the client at the end of the engagement. Hirelyx has not independently audited them.
The challenge
What was actually wrong
The support inbox was dominated by a small set of repeated questions — delivery timing, returns policy, order status, product fit — asked in many different phrasings.
An earlier off-the-shelf chatbot had been removed because it answered confidently from general knowledge instead of the retailer's own policies.
The solution
What we built
A FastAPI service with a retrieval pipeline over the retailer's approved help content and policy pages, so every answer is drawn from their own material and cites where it came from.
Order-status questions are answered through a scoped tool call rather than from text. Anything outside the defined scope, or below the confidence threshold, hands over to a human with the transcript attached.
Implementation
How the work was carried out
The sequence that mattered, including the decisions that prevented later rework.
Technology stack
- OpenAI API
- Python
- RAG
- FastAPI
- Vector database
- 01
Ticket clustering
Historic conversations were grouped to establish what was genuinely repetitive, which set the agent's scope.
- 02
Retrieval over approved content
Help centre and policy pages ingested, chunked and embedded, with a refresh routine when content changes.
- 03
Scoped tooling
A single read-only order-lookup tool with validated input, rather than broad database access.
- 04
Evaluation before launch
A graded test set built from real questions, with accuracy and failure modes reported before rollout.
Outcome
Repetitive queries are now resolved without a queue, and the team's time shifted to the cases that genuinely need a person.
Visuals
Project visuals
Services delivered
What this project drew on
01
Custom AI Agents
Support agents, document processing, lead qualification and multi-step workflows — built on retrieval over your own content, with evaluation and escalation designed in.
02Prompt Engineering
System prompt architecture, structured output contracts and graded evaluation sets — so a prompt change is a measured improvement rather than a hopeful edit.
More work
Other projects
Want something like this for your business?
Tell us where you are now. You will get a written scope, a fixed quote and an honest answer about whether this is the right next step.
- A written scope and fixed quote before work starts
- Payments handled securely through Stripe
- Code, accounts and documentation stay yours
- Typical first reply within 24–48 hours