Skip to Content
Recruiting on behalf of our client

Machine Learning Engineer – On-Premise LLM Agents - Madagascar

Location to be confirmed 06/10/2026

JOB DESCRIPTION

Machine Learning Engineer – On-Premise LLM Agents

Information Technology

Ref.: DIA-2026-IT-18

 

■ Location

■ Seniority

■ Estimated duration

■ Contract

Madagascar – Remote

Senior – 5 to 10 years

6 months to 1 year

To be confirmed

ABOUT THE PROJECT

Our client, an international multi-entity group, is building a security-critical, privacy-first information system on its own on-premise infrastructure: hardened mobile and desktop applications, hardware-key authentication, legally isolated data for each group entity, and self-hosted finance and AI capabilities.

The programme is delivered by a senior, multi-disciplinary team working in two-week agile sprints, with compartmentalised access and a strong culture of security by design.

WORK ARRANGEMENT

This is a remote position: you will work from home, from Madagascar. No attendance at DI-Africa's offices is required.

THE ROLE

Build on-premise LLM agents supporting accounting and finance teams, with strict tool policies, per-entity retrieval and evaluation harnesses, on an air-gapped GPU infrastructure.

KEY RESPONSIBILITIES

●    Deploy quantised open-weight models (7B–32B class) on llama.cpp, with vLLM as fallback.

●    Build accounting and finance agents with hard tool policies and restriction lists.

●    Implement per-entity RAG with strict tenant isolation for AI.

●    Build evaluation harnesses for agent quality and safety.

●    Integrate agents with ERP tools, the ERP remaining the sole caller.

TECHNICAL ENVIRONMENT

●    Python (serving, evaluation, RAG, agents), CUDA literacy, Go / TypeScript integration.

●    llama.cpp, vLLM, Qwen / DeepSeek / Mistral-class models.

●    Air-gapped GPU network, signed model weights.

PROFILE

Location

●    Based in Madagascar, working remotely from home.

Experience

●    Senior – 5 to 10 years of relevant professional experience.

Required skills

●    Production on-premise LLM.

●    Hard tool policies.

●    Evaluation harnesses.

●    Tenant isolation for AI.

Nice to have

●    Quantisation and multi-GPU.

●    Document AI.

Languages

●    Professional English required; French is a plus.

Ways of working

●    Work independently on software development tasks.

●    Write clean, straightforward, efficient code that meets acceptance criteria and specifications.

●    Cover code with unit, property and integration tests where applicable.

●    Write code documentation.

●    Willingness to handle hardware: YubiKeys, HSMs, air-gapped machines, phone flashing.

●    Comfortable working with compartmentalised access.

●    Proactive, requiring minimal supervision or mentoring.

●    Proactive in identifying potential problems and implementing solutions.

●    Efficient time management and respect of deadlines.

●    Practical experience with agile software development methods.

 

HR Consulting · www.di-africa.com

Confidential – For recruitment purposes only

About DI-Africa

Your HR & workforce partner in Africa
14
countries
500+
clients
200+
hires / year

DI-Africa supports industrial, mining, oil & gas and services groups with their hiring across Africa: Employer of Record (EOR), recruitment & sourcing, payroll & compliance, immigration & work permits, international mobility, health insurance.

Our recruitment process

Simple, transparent, supported
  1. Online application — acknowledgement within 48 hours.
  2. DI-Africa interview — conversation with a recruitment consultant.
  3. Client interview — short-list presented to the client.
  4. Offer & onboarding — we support you until your first day.

Is this role for you?

Your application is handled by the DI-Africa Recruitment team.

Apply Now All jobs