Alkmist · Ghent
About Alkmist
Alkmist is a B2B SaaS startup founded in 2024 in Ghent, on a mission to fix broken collaboration workflows in professional services. We replace email-driven chaos with structured, AI-powered Task Flows, started with audit and accountancy, expanding into banking, insurance, legal, M&A, and municipal governance.
We work with international network firms like RSM, Bakertilly, and Nexia, serving 5,000+ users across 64+ countries. We’re ISO 27001 certified, backed by €1.8M in funding.
Our backend is written in Rust. Our team is small, technical, and moves fast. If you’ve ever wanted to shape an AI/ML stack from the ground up inside a company that’s already generating revenue and signing enterprise clients, this is it.
The Opportunity
We’ve just secured a VLAIO R&D subsidy to build the next generation of our AI capabilities. This is a 21-month research & development project spanning six technical work packages, and we need an AI Engineer to drive the machine learning and NLP work at its core.
You won’t be fine-tuning models in isolation. You’ll be building production AI systems that process real financial documents, extract knowledge from meetings and emails, orchestrate multi-agent workflows, and run behavioral experiments, all within a platform that’s already live with paying enterprise customers.
Alongside the VLAIO project, you’ll contribute to our internal program to accelerate Alkmist’s growth trajectory and hit ambitious product-led growth targets.
What You’ll Work On
Document Classification & Intelligent Matching
Build ML models that automatically classify financial documents (depreciation tables, bank statements, trial balances, contracts) and match client uploads to the correct items in a Document Request List. Target: ≥95% classification accuracy, ≥70% contextual matching accuracy, processing folders in under 30 seconds.
AI-Curated Knowledge Base
Design NLP pipelines that extract actionable insights from meeting transcriptions, email threads, and uploaded documents, building a queryable, permission-aware knowledge base per client engagement. The goal: eliminate the repetitive questions caused by 30% annual staff turnover in audit.
Automated DRL Generation
Parse Belgian financial statements and trial balances using OCR + NLP, then auto-generate Document Request Lists with adaptive questioning logic. Target: ≥70% time savings vs. current Excel workflows, validated through crossover experimental design.
Pricing Prediction Engine
Develop ML models (gradient boosting and/or deep learning with categorical embeddings) to replace intuition-based pricing with data-driven predictions. SHAP-based explainability, confidence intervals, and drift monitoring included.
AI Agent Orchestration
Architect and build a multi-agent system where fine-tuned LLMs (Mistral, Qwen, DeepSeek) coordinate specialized agents for document validation, deadline tracking, reconciliation checks, and automated follow-ups. Sub-2-second orchestration decisions, 100+ concurrent users.
Behavioral Nudging & Experimentation
Design and run A/B experiments using multi-armed bandits to optimize client responsiveness, testing reminder timing, message framing, and escalation sequences. Statistically validated improvements in document submission turnaround times.
Your Profile
Required
Nice to Have
What We Offer
How to Apply
Send your CV and a brief note on what excites you about this role to [email protected]. We don’t need a cover letter, a few sentences about a project you’re proud of tells us more.
Questions? Reach out to Toto De Brant (Co-founder) on LinkedIn or at the email above.