RGE Hiring! Full Time AI Implementation Specialist (6 months contract - extendable) in - Ricebowl

AI Implementation Engineer

RGE Digital· Singapore ·6-month contract (extension available)· Start: Flexible

About RGE Digital

RGE Digital is the internal technology engine for RGE Group — a global bio-based resources company with $20B+ in assets and 80,000 employees across four continents. We build and deploy production AI systems across industrial operations, shared services, and corporate functions. Our work spans computer vision for palm oil quality grading, intelligent document processing for financial workflows, and GenAI platforms used daily by hundreds of staff across the group.

We are a commercially-focused, high-output team. Everything we build goes into production.

The Role

We are hiring an AI Implementation Engineer to design, build, and deploy custom AI solutions across RGE's business units. This is an engineering role — you will write code, build pipelines, and ship things that real teams depend on.

Off-the-shelf tools get us part of the way. Your job is to close the gap — building custom solutions where standard products fall short, integrating AI into workflows that have never had it, and making it work reliably in a business environment.

You will be embedded across multiple projects simultaneously: GenAI applications for corporate functions (HR, Finance, Legal), intelligent document processing for shared services, and computer vision for agricultural operations.

What You Will Build

• Custom GenAI pipelines using LLM APIs (Claude, GPT-4) — RAG systems, tool-use agents, structured output extractors, multi-step reasoning workflows
• Intelligent document processing solutions — extracting structured data from invoices, claims, contracts, and forms using layout-aware models and OCR pipelines
• Integrations between AI outputs and business systems (Workday, SharePoint, internal dashboards)
• Evaluation frameworks to measure model accuracy, reliability, and business impact in production

Base Requirements — All Candidates

Every candidate must meet these before we consider track specialisation.

• Python — writes production-quality code — functions, classes, error handling, environment management (venv/conda). Not just notebook scripts.
• LLM fundamentals — understands tokenisation, context windows, temperature, top-p, tool/function calling, system vs user prompts, and why hallucinations happen. Can reason about model behaviour, not just observe it.
• API integration — has called an LLM API (OpenAI, Anthropic, or equivalent) programmatically, parsed structured outputs (JSON mode, Pydantic), and handled rate limits and retries.
• Prompt engineering depth — beyond basic prompting: few-shot, chain-of-thought, output formatting constraints, role prompting. Understands when prompt engineering is the right lever vs. when you need something else.
• RAG basics — understands the retrieval-augmented generation pattern: chunking, embedding, vector search, context injection. Has implemented or studied at least one working example.
• Data handling — comfortable with pandas, JSON/CSV parsing, file I/O, and cleaning messy real-world inputs.
• Version control — uses Git. Commits meaningful units of work. Can explain a diff.
• Debugging discipline — reads stack traces, isolates failures, tests components independently. Does not guess-and-pray.

Specialisation — Pick One Track

We expect every candidate to meet the base requirements above. Beyond that, we are looking for depth in at least one of the following tracks.

Track A — GenAI & Agentic Systems

• Orchestration frameworks — has built pipelines using LangChain, LlamaIndex, LangGraph, or equivalent. Understands chains, agents, tools, memory, and callbacks — not just tutorials, has debugged a broken pipeline.
• Agentic patterns — understands ReAct, tool-use loops, planning and routing. Has built a multi-step agent that calls external tools (search, code execution, database lookup, API) and handles intermediate failures.
• Vector stores and retrieval — has worked with at least one vector database (FAISS, Chroma, Pinecone, Weaviate). Understands embedding models (OpenAI, sentence-transformers), chunking strategies, and retrieval quality evaluation.
• Structured output extraction — extracts specific fields from unstructured text reliably: JSON mode, function calling, regex fallback. Understands precision vs recall trade-offs in extraction tasks.
• Evaluation — has measured LLM pipeline quality: accuracy, hallucination rate, latency, cost-per-query. Has iterated on a pipeline based on evaluation results, not intuition.

Document intelligence (bonus) — experience with layout-aware document models (LayoutLM, Azure Document Intelligence, AWS Textract, Google Document AI). Understands the difference between raw OCR and structured extraction. Has handled multi-format documents (PDF, scanned images, mixed layouts).

Fine-tuning (bonus) — has fine-tuned or supervised fine-tuned (SFT) a model using Hugging Face Transformers, LoRA/QLoRA, or a managed service. Understands when fine-tuning is and isn't the right call vs. prompt engineering or RAG.

Track B — Computer Vision

• Deep learning frameworks — proficient in PyTorch (preferred) or TensorFlow. Writes custom training loops, not just Keras .fit(). Understands backpropagation, loss functions, optimisers, and learning rate schedules.
• Model architectures — familiar with CNN architectures (ResNet, EfficientNet, MobileNet) and detection frameworks (YOLOv8/v9, Faster R-CNN). Understands when to use classification vs detection vs segmentation.
• Transfer learning — has fine-tuned a pretrained model on a custom dataset. Understands feature extraction vs full fine-tuning, layer freezing, and domain adaptation.
• Real-world data handling — has dealt with non-benchmark datasets: class imbalance, inconsistent image quality, missing labels, augmentation strategies (Albumentations, torchvision transforms). Knows the difference between validation accuracy and production performance.
• Model evaluation — goes beyond accuracy: precision, recall, F1, confusion matrix, IoU (for detection). Has analysed failure modes and fed findings back into data or model decisions.
• MLOps basics — has packaged a model for inference: ONNX export, TorchScript, or a REST API wrapper (FastAPI/Flask). Understands batching, latency, and throughput trade-offs.

Multimodal models (bonus) — experience with vision-language models (LLaVA, GPT-4V, Gemini Vision, PaliGemma). Has used a multimodal model for structured data extraction from images.

Edge deployment (bonus) — has deployed a model outside the cloud: Jetson Nano, ONNX Runtime, TensorRT, or equivalent. Understands quantisation and model compression trade-offs.

Who You Are

• Recent graduate (within 2 years) or final-year student in Computer Science, Electrical Engineering, Data Science, or a related field
• Strong academic record — we care about how you think, not just what you know
• Has a portfolio of built things — side projects, hackathons, research implementations. We will ask you to walk us through one.
• Can explain a technical system to a non-technical stakeholder without dumbing it down

What You Will Work On

• HR Claims AI — intelligent processing of 100K+ annual employee expense claims across 10 markets
• AP Invoice OCR — automated extraction and validation of invoice data for Averis shared services
• FFB Quality Detection — computer vision system monitoring fresh fruit bunch quality at palm oil mills across Indonesia
• AI Groundup Initiative — deploying and customising GenAI agents across corporate functions at Group and Business Group level

How to Apply

Send your CV and a short technical note (max 200 words): describe one thing you built using AI — what the problem was, what you built, and what happened when it ran. A GitHub link is welcome.

State which track you are applying for: Track A (GenAI & Agentic Systems) or Track B (Computer Vision).

Shortlisted candidates will complete a short technical screen before a panel interview with the RGE Digital team.