Amharic Support Assistant: Bilingual AI
A sample engagement showing a retrieval-augmented assistant that answers customer questions in Amharic and English using only the company's own documents. When the assistant is unsure, it hands the conversation to a human agent with the context attached.
Amharic + English
Languages
Grounded
Answers
Human
Handover
Illustrative
Figures
Challenge & Scope
Requirements
Answer common questions in Amharic and English with consistent quality
Ground every answer in approved company documents, with sources
Hand difficult or sensitive conversations to a human agent
Let the support team update knowledge without engineering help
Strategic Objectives
Help a support team answer routine questions instantly in the customer's language while keeping humans in charge of the hard cases.
Key Deliverables
Technical Execution
Bilingual Retrieval
Documents are chunked and embedded with models evaluated on Amharic text, and stored in pgvector alongside metadata for filtering.
Guarded Generation
Answers are generated only from retrieved passages, cite their sources, and fall back to a handover when confidence is low.
Continuous Evaluation
A test set of real question types runs on every change, so quality is measured rather than guessed.
Technical Stack
AI
Python
LangChain
Hugging Face
OpenAI
Backend
Python
PostgreSQL
Redis
Frontend
React
TypeScript
Infrastructure
Docker
Google Cloud
Similar Project?
Let's build your next platform together.
Features & Outcomes
Cited Answers
Each reply links back to the document passage it came from.
Smart Handover
Low-confidence chats move to an agent with full context.
Self-Serve Knowledge
Support staff upload and retire documents themselves.
Client Benefits
Engineers experienced with Amharic and Afaan Oromo language data
Measured quality through evaluation sets, not demos
Privacy-aware deployment options, including in-country hosting
Clear documentation so the client team can maintain the system
Strategic Outcomes
Target: faster first responses for routine questions
Designed to keep answers consistent across languages
Support agents focus on complex conversations
A reusable base for more AI features later
System Interfaces
Project Team
Tech Lead
AI Engineer
Retrieval pipeline & evaluation
Engineer
Full-stack Engineer
Chat widget & agent console
Want to talk to an Arki engineer?
Before any commercial conversation, we'll put you on a 30-min call with a senior engineer in your stack. No salespeople. Just a technical sanity check.
Tell us what you're working on.
We'll respond within one business day with the right next step.