( SERVICES ) — AI AGENTS

AI agents grounded
in your own content

One kind of chatbot invents an answer about your refund policy. The other reads the policy. That difference is what separates a demo from something you can put in front of customers.

from $2,000 Get a scoped quote → Read the FAQ
( 01 ) — OVERVIEW

The word chatbot covers three very different products at three very different prices, and most disappointment comes from buying the cheapest and expecting the most expensive. A scripted bot follows a decision tree. A raw language model answers confidently from general knowledge, including about your business, which it does not have. A grounded agent retrieves your actual documents and answers from those.

Only the third is safe to point at customers. Grounding — retrieval-augmented generation — means the agent looks up your policies, your pricing, your intake forms and your FAQs before answering, and cites what it found. When it does not find an answer, it says so and hands over to a person rather than improvising.

Guardrails matter as much as capability, particularly in regulated or high-stakes contexts. The agent must know what it must never answer, and every one of those cases needs a clean human escalation path. We define that boundary explicitly during scoping, because the failure mode of a confident wrong answer is worse than no answer at all.

Build cost and running cost are two different numbers, and conflating them is the most common budgeting surprise. An agent has a one-off build and an ongoing per-conversation cost that scales with usage. We quote both, because only knowing the first is how projects get abandoned in month three.

( 02 ) — WHAT'S INCLUDED

Everything in
the from $2,000.

Fixed price, fixed scope, agreed in writing before anything starts. You own everything on day one.

01
Use-case and boundary scoping

What the agent handles, and — just as important — what it must always escalate to a person.

02
Knowledge base preparation

Your policies, FAQs, product data and documents cleaned and structured for retrieval. Most of the quality comes from this step.

03
Retrieval-grounded architecture

pgvector-backed retrieval so answers come from your documents rather than the model's general knowledge.

04
Model selection

Anthropic Claude or OpenAI chosen for the job, balancing answer quality against per-conversation running cost.

05
Escalation and handover

A clean path to a human with the conversation context attached, so the customer does not start over.

06
Channel deployment

On your site, on WhatsApp, or both — whichever channel your customers already use.

07
Guardrails and testing

Adversarial testing against the cases the agent must refuse, before it goes anywhere near a customer.

08
Cost modelling and monitoring

Build cost and projected running cost quoted separately, with usage monitoring after launch.

* Indicative starting prices in USD. Final quotes scoped per project — VAT handled for UAE & EU clients.

( 03 ) — WHO THIS IS FOR

Right fit,
and honest misfits.

An agent earns its cost when the same questions arrive repeatedly and the answers genuinely exist in writing somewhere.

Businesses answering the same set of questions every day across phone, email and WhatsApp
Clinics, retailers and service businesses whose front desk is the bottleneck at peak hours
E-commerce sellers where order status and returns dominate support volume
Teams with documented policies that customers never manage to find on the site

We will talk you out of it in these cases:

×The answers are not written down anywhere — an agent cannot ground itself in knowledge that only exists in someone's head
×Every enquiry genuinely needs human judgement, which some professional services do
×The volume is low enough that a person answering is simply cheaper than a build plus running costs
( 04 ) — HOW IT RUNS

Four steps,
no surprises.

01Scope
Use cases, refusal boundaries and escalation paths agreed in writing before anything is built.
02Ground
Documents prepared and indexed for retrieval — the step that determines answer quality.
03Build & test
Agent built, then tested adversarially against the cases it must refuse.
04Deploy & monitor
Live on your channels, with usage and cost monitoring from day one.
( 05 ) — RELEVANT WORK
Mr. PA AI CONCIERGE · OPERATIONS AUTOMATION — 2025

A solo founder running two agencies at once, with calendars clashing and vendors slipping through the cracks. We built a concierge platform that syncs calendars, budgets and vendors across both agencies with adaptive AI workflows — one console, two businesses.

Mr. PA runs adaptive AI workflows across two agencies from a single console — an agent doing real operational work rather than answering questions.

Read the full case study →
FURTHER READING
( 06 ) — FAQ

AI Agents,
answered straight.

How much does an AI agent cost?

AI agents start from $2,000 to build, an indicative starting price in USD. Separately from the build, there is a running cost per conversation that scales with usage. We quote both numbers, because budgeting only for the build is the most common way these projects get abandoned a few months in.

What is a RAG agent and why does it matter?

RAG — retrieval-augmented generation — means the agent retrieves your actual documents before answering, rather than generating from general knowledge. It is the difference between an agent that invents an answer about your refund policy and one that reads the policy. Only the grounded kind is safe to put in front of customers.

Will it make things up?

Grounding plus explicit guardrails is how you prevent that. The agent answers from your documents, and where it cannot find an answer it says so and escalates to a person. We test adversarially against the cases it must refuse before it goes live — that testing is part of the build, not an extra.

Can it work on WhatsApp?

Yes. For a lot of businesses in this region WhatsApp is not a channel but the primary sales and support interface, so we deploy there as readily as on a website — and often to both.

Can it handle Tamil?

It can, and whether it should is a real question rather than an automatic yes. For consumer-facing businesses a meaningful share of customers are more comfortable typing in Tamil, particularly older and first-time online buyers. We will give you an honest read on whether it is worth it for your specific audience.

What happens when the agent cannot answer?

It hands over to a person with the conversation context attached, so the customer does not have to explain themselves twice. Defining exactly when that handover triggers is part of scoping, and in regulated contexts it is the most important part.

( 07 ) — OTHER SERVICES
Web Design from $1,200 Landing Pages from $800 Web Applications from $4,500 SEO, AEO & GEO from $600/mo Marketing Funnels from $1,000 System Automation from $1,500 MVP Building from $7,500
← All services & pricing