Berlin based - m/f/d - full-time

What we do at Plato

Plato is the AI-powered Sales Intelligence platform built specifically for B2B wholesale. We transform reactive sales teams into proactive revenue engines by surfacing hidden opportunities in their data.

🔍 The Challenge

Why this role is different:

B2B wholesale sits on massive amounts of data — millions of transactions, complex product catalogs, customer communications — but accessing it requires SQL expertise or clicking through endless dashboards. We're building AI that lets users simply ask questions and get answers, extract information from documents automatically, and find what they need instantly.

This isn't consumer AI. This is enterprise language intelligence for wholesale.

In B2C, you optimize for engagement. In B2B wholesale, you optimize for precision and trust. Your text-to-SQL needs to handle complex joins across orders, customers, and products. Your search needs to understand that "M8 hex bolt DIN 933" and "hexagon head screw 8mm" might be the same thing. Your extraction pipeline needs to parse a PDF quote request and map it to 50,000+ SKUs without errors.

Your systems will power real business outcomes:

A user searches for a product and finds it even when their terminology doesn't match the catalog
A sales rep asks "Which customers decreased orders by more than 20% this quarter?" and gets an instant answer from their data
An email arrives with a quote request — your system extracts line items and creates a draft order automatically
A manager asks "What's our revenue by product category for customer X?" and gets a chart, not a ticket to the BI team

Welcome to language AI for wholesale — where natural language meets structured data, and where the right query at the right time transforms how businesses operate.

🎯 Why This Role Matters

Language understanding is becoming core to Plato's value proposition. We're building AI features that require deep NLP capabilities: natural language interfaces to business data, search that understands wholesale terminology, and document extraction that automates manual workflows.

You won't be integrating off-the-shelf solutions. You'll be building language AI systems that become a competitive advantage — text-to-SQL that handles complex B2B queries, domain-specific retrieval, and intelligent document processing.

Short Description

We're looking for a Senior ML/AI Engineer to own language AI at Plato. You'll build the NLP infrastructure that powers multiple product areas: conversational analytics (chat with your data / text-to-SQL), semantic search, document extraction, and AI chat interfaces. This is a broad role combining structured data querying, retrieval systems, and information extraction.

Responsibilities

Build text-to-SQL systems that let users query business data in natural language
Design and build semantic search infrastructure serving both UI search and agentic AI features
Develop information extraction pipelines that turn emails, PDFs, and documents into structured, actionable data
Create AI chat interfaces that combine data queries with document retrieval
Own retrieval architecture: embedding pipelines, hybrid search, reranking, evaluation
Build evaluation frameworks and feedback loops that improve systems over time
Fine-tune models on B2B wholesale domain data where off-the-shelf solutions fall short

🚀 What you'll be working on

Search & Retrieval

Finding relevant information across products, documents, and business data.

Build hybrid search combining BM25 (exact part numbers) with semantic retrieval
Design retrieval APIs serving both UI search (low latency) and agent context (high recall)
Implement multi-tenant search across 20+ enterprise customers
Create domain-adapted retrieval that understands wholesale terminology

Information Extraction & Document Understanding

Turning unstructured documents into structured data that powers automated workflows.

Develop pipelines that extract structured data from emails, PDFs, and quote requests
Build entity extraction for B2B domain: product references, quantities, part numbers, standards
Create automated workflows — e.g., quote request arrives, system extracts line items, creates draft order
Map unstructured text to product catalogs with 500K+ SKUs

End-to-End Ownership

Own architecture decisions — you'll influence the tech stack
Build evaluation frameworks connecting model performance to business outcomes
Design feedback loops: user behavior → evaluation → model improvement
Ship fast, then iterate based on real usage

🏅 What makes you successful here

Technical Foundation

Strong NLP fundamentals — you understand transformers, embeddings, attention mechanisms, not just API calls
Experience building production systems in at least one of: text-to-SQL, search/retrieval, information extraction, or conversational AI
Familiarity with LLMs beyond prompting — RAG architectures, context management, grounding, evaluation
Solid software engineering practices — your code is tested, documented, maintainable

Problem-Solving Mindset

You approach problems from first principles rather than reaching for standard solutions
Comfortable with ambiguity — you can define success metrics when requirements are vague
You balance research exploration with shipping pragmatism
Experience translating business problems into NLP solutions

Proven Track Record In:

Building NLP/ML systems in production (not just notebooks)
Working with LLMs and embedding models beyond basic prompting
Evaluating and improving system quality systematically
At least one of: text-to-SQL, search/retrieval, information extraction, or conversational AI

Bonus Points For:

Experience with text-to-SQL systems, semantic parsing, or natural language interfaces to databases
Knowledge of retrieval systems (hybrid search, reranking, evaluation metrics)
Experience fine-tuning models (embeddings, extractors, or similar) on domain data
Familiarity with OCR, PDF parsing, or document understanding
Understanding of B2B/wholesale domain or ERP systems
Multi-lingual NLP experience (German/English)

🛠️ Our Tech Stack

You'll have significant influence over the NLP/retrieval stack. Current foundation:

Data Processing: PySpark, SQL, Python
ML Platform: Databricks (Unity Catalog, Workflows, Model Serving)
Infrastructure: AWS, Terraform
Orchestration: Databricks Workflows, Github Actions

You'll own decisions on:

Search infrastructure (OpenSearch, Elasticsearch, or alternatives)
Vector store (pgvector, Qdrant, Weaviate, Databricks Vector Search)
Text-to-SQL approach (fine-tuned models, prompted LLMs, hybrid)
Document processing pipeline (OCR, parsing, extraction)

🔥 How We Work

High ownership, low process. You own problems end-to-end — from architecture to production. No handholding, no lengthy specs, no bureaucracy.

Ship fast, iterate faster. The system generating value today beats the perfect one still in notebooks. We default to speed — make the call, ship v0, learn, improve.

Direct and intense. Walk over to product. Call the customer. Jump into the code. This is high-intensity work building something hard.

Small team, big impact. We're intentionally small — every person raises the bar. Your systems power core product features, your decisions shape the platform.

📍 Work Environment

Modern office in Berlin-Mitte with a collaborative team culture
Small, elite team where everyone knows everyone
Direct access to founders and customers — no bureaucracy
Your work ships fast and impacts the business immediately

📝 Interview Process

Initial conversation about your experience and interests
Technical deep-dive on NLP systems you've built
Architecture case study: design a language AI feature for B2B wholesale
Meet the broader team and founders

We are excited to hear from you!

Miguel

Compensation: €90K – €130K • Offers Equity

• €90K – €130K • Offers Equity