Prompt /AI Engineer (LLM, RAG& Fine‑Tuning)

KiteSense Pte Ltd; Singapore; Hybrid; SGD 60,000 - 100,000; Full time

Job summary

Join a forward-thinking company dedicated to transforming education through AI. As a Prompt/AI Engineer, you'll design and optimize prompts, build retrieval-augmented generation systems, and fine-tune models to enhance learning outcomes for thousands of children. This role offers the unique opportunity to directly impact how students engage with AI tutors, fostering a growth mindset and improving exam results. With a flexible remote-friendly environment, you'll thrive in a collaborative setting that encourages innovation and rapid iteration. If you're passionate about leveraging technology to shape the future of education, this is the perfect opportunity for you.

Benefits

Mission & Impact

Ownership

Flexibility

Qualifications

• 3+ years building production NLP or LLM systems.
• Deep Python skills with orchestration stacks.
• Experience with fine-tuning techniques.

Responsibilities

• Design and optimize prompts for AI tutoring.
• Own the RAG pipeline and measure performance.
• Collaborate with SMEs to improve AI marking.

Skills

NLP systems

Python

vector databases

fine-tuning

prompt engineering

Tools

LangChain

LlamaIndex

Haystack

W&B

MLflow

Job description

Prompt /AI Engineer (LLM, RAG& Fine‑Tuning) Full‑time · Singapore HQ · Remote‑friendly (GMT+5to+11)

AboutKiteSense

KiteSense gives every child a caring, world‑class AI tutor. Our platform fuses large‑language‑models, adaptive learning paths, and MOE‑aligned content to lift exam results and ignite a growth mindset. We already serve over thirty thousand learners and we’re scaling fast across SoutheastAsia.

Why This Role Matters

The prompts, retrieval stack, and fine‑tuned adapters you build determine how CoachKyu thinks: his explanations, marking accuracy, motivation style, and even cost‑per‑session. A single tweak can cut hallucinations, drive retention, and delight parents overnight. You’ll hold the keys to that engine.

What You’ll Do

30% - Design & Optimise Prompts

Example: Craft version‑controlled templates for marking, feedback, coaching, motivation. Use CoT, JSON mode, function calling, and hybrids (few‑shot+RAG) to eliminate marking error.

25% - Own the RAG Pipeline

Example: Build and iterate retrieval‑augmented generation: embeddings, vector DB (PGVector/Pinecone), hybrid BM25+dense search, semantic caching, re‑rankers (e.g. ColBERT). Measure precision@k, answer grounding, latency.

20% - Domain Fine‑Tuning & Adapters

Example: Train LoRA/QLoRA adapters or continual‑pretrain on MOE‑aligned corpora (past PSLE papers, student chat). Track experiments (W&B/MLflow), guard against drift and over‑fit.

15% - Evaluation, Observability & Safety

Example: Extend automated harnesses for accuracy, tone, bias, hallucination. Ship real‑time drift alerts and rollback scripts. Maintain policy compliance & privacy safeguards.

10% - Collaboration & R&D

Example: Run “Prompt Clinics” with Curriculum SMEs & Academic Director. Demo quarterly moon‑shots (multimodal retrieval, RLHF on student up‑votes, teacher‑forcing loops).

You’ll Thrive If You Have

• 3+yrs building production NLP or LLM systems (OpenAI/Anthropic/Cohere or open‑source).

• Deep Python plus one orchestration stack (LangChain, LlamaIndex, Haystack).

• Hands‑on experience with vector databases and retrieval evaluation.

• Practical fine‑tuning chops: LoRA/QLoRA, PEFT, experiment tracking

• Ability to translate pedagogical goals into precise, testable prompt/RAG specs.

• Growth mindset—iterate fast, love shipping weekly, and learn from data.

Bonus Points

• K‑12 assessment or tutoring‑tech background.

• Experience with RLHF, reward modeling, or policy‑tuning.

• Familiarity with OpenMetadata/Marquez, DataDog, or similar observability stacks.

• Multilingual prompt/RAG work (Mandarin, Bahasa Indonesia, etc.).

What We Offer

• Mission & Impact – Shape how thousands of kids learn every day.

• Ownership – Production access; your work hit users in hours.

• Flexibility – Hybrid Singapore HQ or remote within GMT+5 to+11.

How to Apply

Email [e-mail] with:

1. Your CV / LinkedIn.

2. A short case study of a tricky LLM prompt, RAG tweak, or fine‑tune you led (context & outcome).

3. In ≤150words, one idea to improve AI marking accuracy for Singapore PSLE questions.

We review weekly and aim to reply within7days. Come build the brain behind the next generation of AI tutors!