Job Responsibilities
1. Product Strategy & Planning
- Own product roadmap planning for [specific product domain, e.g., AI customer service/content generation tools/data analysis Copilot].
- Identify "AI Native" opportunities aligned with company strategy and market trends; deliver PRDs (Product Requirement Documents) and business case reports.
- Analyze competitor AI product capabilities (e.g., compare generative quality, response speed, pricing strategies across models).
2. AI Capability Definition & Implementation
- Requirements Translation: Convert vague business needs (e.g., "make responses more human-like") into clear, algorithm-team-evaluable technical specs (e.g., "support multi-turn dialogue with coreference resolution; emotion recognition accuracy must reach 85%").
- Strategy Design: Design human-AI interaction strategies (e.g., fallback logic, confidence thresholds, active clarification mechanisms); handle edge cases from model uncertainty.
- Prompt Engineering & Fine-tuning: Lead or collaborate on prompt optimization, few-shot example design; participate in setting data annotation standards for SFT (Supervised Fine-Tuning).
3. Model Performance Evaluation & Iteration
- Establish proprietary evaluation systems for AI products (beyond functional usability: accuracy, recall, hallucination rate, response latency, user satisfaction).
- Design A/B testing frameworks comparing model versions/strategies against business metrics (e.g., conversion lift, customer service resolution rate).
- Drive badcase analysis: establish closed-loop process of "data annotation → model retraining → effect validation."
4. Cross-functional Collaboration
- With Algorithm Team: Define model performance metrics (inference latency, VRAM usage); assist in training/inference resource budgeting.
- With Development Team: Define front-end/back-end interaction protocols (API design); handle async model invocation, caching strategies, and degradation plans.
- With Data Operations: Set data collection and annotation standards (especially for high-subjectivity tasks like "is the response helpful").
Qualifications
Basic Requirements
- Bachelor's degree or higher in Computer Science, Mathematics, Statistics, AI, or related field (Master's preferred).
- 5+ years B2B/C2C product experience, with at least 2+ years dedicated AI product experience (dialogue systems, image recognition, recommendation engines, AIGC applications, etc.).
Professional Skills
- Technical Understanding:
- Grasp fundamentals of machine learning/deep learning (supervised/unsupervised/reinforcement learning); distinguish discriminative vs. generative models.
- Familiar with common AI frameworks: Transformer architecture, RAG (Retrieval-Augmented Generation), Agent, LoRA fine-tuning, etc.
- Must-have: Basic SQL proficiency (data exploration); able to read simple Python pseudocode (API parameter discussion).
- Evaluation Methods: Master blind testing and human evaluation design; familiar with NLP metrics (BLEU, ROUGE, accuracy, F1) and their limitations.
- Tools: Proficient with at least one AI prototyping tool (LangSmith, Dify, Coze, etc.); able to independently build demos for feasibility validation.
Cognitive Abilities
- Non-deterministic Thinking: Can design experiences for "probabilistic products" (accept model errors; pre-design correction, feedback, degradation flows).
- Data Intuition: Abstract model capability gaps from badcases rather than simply blaming "model dumbness."
- Cost Awareness: Understand tradeoffs between models (GPT-4 vs GPT-3.5 vs proprietary models) on performance, latency, and cost.
Bonus
- Hit AIGC Product Experience: Participated in AIGC products with 100k+ DAU or launched commercialized AI features.
- Technical Background: Career switcher (e.g., algorithm engineer → PM) or independently able to call OpenAI APIs/open-source models (Llama, Qwen) for batch testing scripts.
- Industry Depth: Deep domain expertise [law, healthcare, programming, education, etc.] + AI implementation experience in specific verticals.
- Paper Sense: Follow AI frontier (Arxiv daily); quickly assess new papers' impact on current products (e.g., how long-context tech changes RAG design).
- Obsessive AI Passion: Heavily use AI tools daily (ChatGPT, Midjourney, Copilot); built personal workflow automation.