Job details:

• Start date: ASAP
• HackerRank Challenge: Yes
• Duration: 1 year
• Remote vs Onsite: Fully remote, occasional in-person team sessions/workshops/gatherings (~1x/quarter) likely in Prague. Candidate who can work from prague office will be given advantage
• US Hours overlap needed: Minimum 2–6pm CET, preferred 2–7pm CET

Project overview:

This project will build a platform that leverages AI-powered document ingestion, retrieval, and reasoning by combining RAG, search, and document drafting. The goal is to help teams accelerate analysis, improve consistency, and reduce reliance on manual review. The result will be a more repeatable, auditable, and scalable approach to consolidation scope assessments, reducing effort and cost for clients while enabling client to deliver insights faster and at greater scale.
 

About the role:

We’re looking for a Python Engineer with strong RAG/LLM engineering experience and supporting Node.js/React skills to build and iterate on a fast-moving internal platform and demo toolchain. This role is hands-on across:

• Python services (FastAPI) and async processing

• Hybrid retrieval (vector + BM25) with Azure AI Search (ACS / Azure Cognitive Search) and/or Qdrant

• Document pipelines (PDF extraction, OCR, chunking, metadata filtering)

• Structured output generation (Excel/Word memos) and demos for non-technical stakeholders

• Maintaining integration modules in Node.js/React that connect the frontend to Python APIs

The solution is rapid iteration, prompt versions, evolving requirements, but we expect strong engineering hygiene (tests, commits, Docker, reproducibility).
 

Scope:

Fullstack & Platform Engineering (Core)

• Build and maintain backend APIs in Python (FastAPI) with clean architecture, strong typing, documentation, and reliability.

• Design and implement asynchronous background processing using Celery and messaging patterns.

• Work with PostgreSQL and MongoDB where appropriate to design data models for document metadata, runs, outputs, and configuration state.

• Implement event-driven workflows using Pub/Sub patterns and messaging tech such as RabbitMQ and Azure Service Bus is a plus.

RAG / Retrieval / LLM Engineering (Core)

• Implement end-to-end RAG pipelines: ingestion → chunking → embeddings → retrieval → generation → structured outputs.

• Build hybrid retrieval patterns combining:

• vector search (embeddings) + BM25 sparse retrieval

• filtering by metadata/pool (collections, tags, entity IDs)

• Integrate and tune retrieval with Azure AI Search (ACS) and/or pgVector, Qdrant (collections, filters, indexing lifecycle).

• Implement and maintain prompt engineering workflows with iterative prompt versioning including structured output parsing and guardrails.

• Use OpenAI API (reasoning models + structured outputs) and embedding models as part of production-like pipelines.

Document Processing & Knowledge Ingestion

• Build robust ingestion for legal/financial documents:

• PDF text extraction (PyMuPDF/fitz)

• OCR for scanned documents (Tesseract/pytesseract)

• clause-aware / section-aware chunking strategies for better retrieval quality

Frontend / Demo Enablement

• Fast PoC by building and maintaining Streamlit multi-page internal tools (session state, interactive grids like streamlit-aggrid) to support demos and SME workflows.

Fullstack Integration & Maintenance

• Update and maintain the Node.js + React service, especially integration modules that connect the frontend to Python backend APIs, ensuring compatibility and end-to-end reliability.

• Contribute to Node.js backend components when required, integrating them with Python services.

Quality & Delivery

• Write and maintain unit/integration tests for Python services and pipelines.

• Containerize services using Docker and Docker Compose (e.g., multi-container Streamlit + Qdrant setups).

• Drive reproducible config using YAML-driven configuration (e.g., configs/, prompts/manifest.yaml) and strong Git hygiene (small regular commits, code reviews, CI/CD).

Requirements:

Python & Engineering Fundamentals

• Python 3.12+, strong OOP, clean architecture.

• CLI scripting and orchestration (argparse, bash).

• Package management, virtual environments, Docker.

Backend

• FastAPI (API design, service-oriented patterns).

• Asynchronous processing: Celery, messaging, event-driven design (RabbitMQ / Azure Service Bus).

• Relational DB experience, preferably PostgreSQL.

RAG / NLP / LLM

• RAG architecture and implementation (chunking, embeddings, retrieval, generation).

• Prompt engineering with versioning and iteration (multi-version prompt lifecycle).

• OpenAI API experience including structured output parsing.

• Embeddings and retrieval tuning; hybrid retrieval approach (vector + BM25).

Vector & Search

• Qdrant (or Pinecone/Weaviate/Milvus) — collection management, metadata filters.

• Azure AI Search / Azure Cognitive Search (ACS) integration for indexing and retrieval.

Document Processing

• PDF extraction: PyMuPDF (fitz).

• OCR: Tesseract/pytesseract.

• Chunking strategies optimized for legal/financial docs (clause/section aware).

Internal Tools / Demo

• Streamlit multi-page apps, session state, interactive grids (streamlit-aggrid).

Data & Reporting

• pandas, NumPy, openpyxl — Excel report generation from JSON.

• Batch processing across multiple entities/projects.

Integration / Fullstack Support

• Basic-to-intermediate Node.js and React (maintain/integrate existing services).

Soft skills:

• Strong communication and ability to work closely with accounting/audit SMEs to refine prompts and outputs (rapid iteration, POC environment).

• Able to debug end-to-end pipelines independently (ingestion → retrieval → generation → output).

• Strong analytical thinking and creative problem-solving.

Nice to have:

• RAG evaluation/scoring frameworks; quality measurement and regression testing for retrieval + generation.

• MCP / FastMCP, LangChain, LangGraph, multi-agent frameworks (CrewAI, AutoGen, TaskWeave).

• Azure CI/CD pipelines and cloud-native deployments.

SNI sp. z o.o. will process personal data for the purpose of the recruitment process in accordance with Data Privacy Policy. The data may also be stored and processed for future recruitment purposes, in accordance with the given consent.