Service · AI & automation

AI chatbots and automation for Ethiopian businesses

I build AI assistants that answer customers and staff in Amharic and English, on Telegram, the web or inside your own systems, and automations that move data between the tools you already use. Five are in daily use today, in pharmacy, HR, ERP, supplier and registration work. None of them is allowed to write its own SQL: each answers only through approved, permission-scoped queries.

Amharic and English
Telegram and web chat
Reads photos and voice
Permission-scoped data access
n8n · Gemini · FastAPI

Message me on Telegram Email +251 703 476 023

What I build

Assistants that do real work

Customer-facing assistants

Chatbots that answer stock, price and order questions from your live data, read photos such as prescriptions, and alert your team when something needs a person, like a stock-out.

Internal assistants for staff

Ask the ERP, HR or reporting system in plain language. Each person sees only what their role allows: identity comes from a signed JWT, and the assistant runs read-only, pre-approved queries.

Amharic voice and text

Voice registration in Amharic that turns speech into structured records, and assistants that reply in Amharic, a language most off-the-shelf bots overlook.

Workflow automation

n8n workflows that sync CRM records to Jira, send Slack or Telegram alerts, process webhooks and schedule reports, so routine work happens without anyone copying data by hand.

Proof

Assistants in daily use

26 nodes · 5 agents

AI pharmacy assistant

Five specialist Gemini agents with shared memory: answers stock questions, reads prescription photos, guides patients to the right medicine and alerts the sales team about stock-outs.

Role-scoped data

HR performance assistant

Employees, managers and HR each see only what their role allows, enforced by FastAPI middleware that verifies a signed JWT before the agent runs a query.

8 live SQL tools

Odoo ERP assistant

Natural-language questions answered from live Odoo data, with runtime schema inspection.

Amharic voice → data

Amharic voice registration

A Gemini 2.5 Flash pipeline: spoken Amharic registration in, structured records out.

Who imports what

Market intelligence chatbot

Answers who imports what, and at what price, from regulator import-permit data loaded by a Python ETL into PostgreSQL.

Process

From idea to an assistant people use

1
Pick one job
We start with one task that costs your team hours every week, such as answering the same customer questions or pulling the same report.
2
Connect your data safely
The assistant gets read-only, permission-scoped access to exactly the data that job needs, and nothing else.
3
Test with real questions
We test with questions your customers and staff actually ask, in Amharic and English, before anyone outside the team sees it.
4
Launch and improve
It goes live on Telegram, your website or your internal tools, and I tune it from real conversations.

Local fit

Built for how Ethiopia works

▹Telegram is where Ethiopian customers already are, so an assistant can live there instead of in an app nobody installs.
▹Amharic in and Amharic out, in text and voice, with Ge'ez script handled properly.
▹Ethiopian names matched across spellings with habesha-names, my open-source library: Ge'ez to Latin and sound-alike matching.
▹Your data stays under your rules: assistants run approved queries only, never free-form SQL against your database.

FAQ

Questions about AI assistants

Can an AI chatbot speak Amharic?

Yes. Modern models such as Gemini read and write Amharic and can transcribe Amharic speech. I have shipped an Amharic voice registration pipeline on Gemini 2.5 Flash, and assistants that reply in Amharic.

Can the bot run on Telegram?

Yes. Telegram bots are one of the channels I build on, alongside web chat and internal tools, and one assistant can serve more than one channel.

Is it safe to connect an AI to our database?

It is when the AI never writes its own queries. My assistants call a fixed set of approved, read-only, permission-scoped queries, and every request carries the user's verified identity, so the AI only sees what that person is allowed to see.

What does it run on?

Usually n8n for the workflow, Gemini, OpenAI or Claude for the model, and a FastAPI service in front of your data, deployed with Docker on your server or a VPS.

Do you build automations without AI?

Yes. Much of the value is plain automation: syncing CRM records to Jira, Slack and Telegram alerts, webhooks and scheduled reports in n8n.

More services

Contact

Which job should an assistant take off your team?

Describe the questions your customers or staff ask most and where the answers live. I will tell you what an assistant for it looks like.