AI Engineer – Agent Reasoning & Multimodal Dialogue

SHOPEE IP SINGAPORE PRIVATE LIMITED; Singapore; On-site; SGD 120,000 - 180,000; Full time

Job summary

SHOPEE IP SINGAPORE PRIVATE LIMITED is seeking a researcher to design and develop core agent algorithms, including multi-turn dialogue planning and tool orchestration, with memory graphs for personalized interactions.

You will advance emotion-aware dialogue modeling, LLM alignment and safety strategies (SFT and DPO), and multimodal reasoning integrating vision with dialogue planning for tutoring or adaptive guidance.

Qualifications

• Hands-on research and engineering experience building conversational agents with memory systems.
• Experience with persona-aware and emotion-aware dialogue modeling and LLM alignment (SFT, DPO).
• Experience orchestrating agent pipelines with retrieval, ranking, and tool use.
• First-author publications at top-tier venues on persona-driven dialogue generation.
• Experience building vision-language tutoring/dialogue agents with multimodal reasoning and reinforcement learning for sequential decisions and 3D reconstruction from point clouds.
• Strong Python and Bash; PyTorch and Hugging Face; LoRA/PEFT; prompt engineering; model evaluation pipelines.
• Experience deploying LLMs/VLMs on AWS/GPU clusters; familiar with PostgreSQL, Neo4j, AWS S3.
• Strong problem analysis and ability to work independently and with teams.

Responsibilities

• Design and develop core agent algorithms including multi-turn dialogue planning and tool orchestration.
• Build and maintain agent memory architectures using knowledge graphs.
• Develop emotion-aware dialogue modeling techniques.
• Design and implement LLM alignment and safety strategies (SFT, DPO).
• Build multimodal agent capabilities that integrate vision-language reasoning with dialogue planning.
• Benchmark and deploy LLMs/VLMs across GPU clusters and cloud environments; build evaluation pipelines.
• Track frontier research and represent findings at AI/NLP venues.

Skills

Python

Bash

PyTorch

Hugging Face

LoRA/PEFT

Prompt engineering

Model evaluation

PostgreSQL

Neo4j

AWS S3

research skills

independence

Education

Master's degree or above in CS/AI/NLP

Tools

PostgreSQL

Neo4j

AWS

GPU clusters

Job description

Job Description:

• Design and develop core agent algorithms, including multi-turn dialogue planning, tool orchestration (retrieval, ranking, LLM synthesis), and adaptive task planning.

• Build and maintain agent memory architectures using knowledge graphs to support long-term consistency, personalization, and context retention across sessions.

• Develop emotion-aware dialogue modeling techniques to improve agent naturalness, consistency, and user engagement.

• Design and implement LLM alignment and safety strategies (e.g., SFT, DPO) to mitigate risks such as implicit persuasion or psychological manipulation in personalized generation.

• Build multimodal agent capabilities that integrate vision-language reasoning with dialogue planning for tasks such as tutoring or adaptive guidance.

• Benchmark and deploy LLMs/VLMs across GPU clusters and cloud environments (e.g., AWS); build reproducible evaluation pipelines to support model and architecture selection.

• Track frontier research in agent algorithms, contribute to publications, and represent findings at top-tier AI/NLP venues.

Requirements:

• Master's degree or above in Computer Science, Natural Language Processing, Artificial Intelligence, or a related field.

• Minimum 3 years of hands-on research and engineering full-time working experience building conversational agents with knowledge-graph-based memory systems, persona-aware and emotion-aware dialogue modeling, and LLM alignment techniques (SFT and DPO) for safety and behavior control, combined with experience orchestrating agent pipelines involving retrieval, ranking, and tool use.

• First-author publication(s) at top-tier venues (ACL/AAAI/EMNLP/ICLR) on persona-driven dialogue generation and persona attribute extraction, particularly methods that improve dialogue consistency and personalization quality.

• Demonstrated experience building vision-language tutoring/dialogue agents that integrate multimodal reasoning with adaptive dialogue planning, applying reinforcement learning (e.g., Deep Q-Network) to sequential decision-making problems, and developing end-to-end 3D reconstruction pipelines (segmentation, planar extraction, geometric reconstruction) from point cloud data.

• Good programming skills in Python and Bash; proficient in PyTorch and Hugging Face; familiar with LoRA/PEFT, prompt engineering, and model evaluation pipelines.

• Experience benchmarking and deploying LLMs/VLMs across GPU clusters and cloud platforms (e.g., AWS EC2); familiar with data systems such as PostgreSQL, Neo4j, and AWS S3.

• Good problem analysis and research skills; sustained curiosity in frontier AI/agent research; able to work independently and collaboratively across research and engineering teams.