Upwork home
Sign up
Project Catalog
DesignDesign
Video & AudioVideo & Audio
Development & ITDevelopment & IT
MarketingMarketing
Writing & TranslationWriting & Translation
MoreMore

You will get a quality audit of your AI agent's real conversations, with a fix list

Zengyu Z.Status: Offline
Zengyu Z. Zengyu Z.
Preview image

Let a pro handle the details

Buy Generative AI services from Zengyu, priced and ready to go.
Zengyu Z.Status: Offline
Zengyu Z. Zengyu Z.
Preview image

Let a pro handle the details

Buy Generative AI services from Zengyu, priced and ready to go.

Project details

Your AI agent is live. Users are talking to it. Do you actually know where it fails?

I will read your agent's real transcripts end-to-end and deliver: per-conversation scoring on a rubric we agree upfront (task completion, answer quality, tone, hallucination, escalation handling), a badcase taxonomy where every failure is classified by type - not just flagged, root-cause clustering that separates retrieval issues from prompting issues from product-design issues, and a prioritized fix list so your team knows what to repair first for maximum quality gain.

The report is written for forwarding: your PM can send it to leadership as-is. Higher tiers add root-cause analysis, an exec-ready summary, and written Q&A follow-up.

This is the discipline used inside large AI product teams - I run evaluation for a production LLM copilot at scale - applied to your data. Bilingual: I audit English, Chinese, and mixed-language conversations.
AI Algorithms
Large Language Model, Multimodal Large Language Model, Transformer Model
AI Applications
AI Chatbot, Conversational AI
AI Models
ChatGPT, GPT-4
What's included
Service Tiers Starter
$500
Standard
$900
Advanced
$1,500
Delivery Time 7 days 10 days 14 days
Number of Revisions
122
AI Model Integration
-
-
-
Batch Normalization
-
-
-
Database Integration
-
-
-
Detailed Code Comments
-
-
-
Image Upscaling
MLOps
-
-
-
Model Deployment
-
-
-
Model Documentation
-
-
-
Model Monitoring
-
-
-
Model Testing & Optimization
-
-
-
Model Tuning
-
-
-
Natural Language Processing
-
-
-
NLP Tokenization
-
-
-
Pre-Training
-
-
-
Prompt Engineering
-
-
-
Setup File
-
-
-
Source Code
-
-
-
Zengyu Z.Status: Offline

About Zengyu

Zengyu Z.Status: Offline
LLM Evaluation & Annotation Specialist | Bilingual EN-CN
Sunnyvale, United States - 10:29 pm local time
I own the quality bar for three shipped LLM agent products at a Global Technical Company - defining what "great" looks like as written rubrics, running the end-to-end human evaluation pipeline behind them (annotator guidelines, golden sets, calibration, disagreement arbitration), and building the automated judges validated against human raters that make quality continuously measurable. My work turns "this answer feels wrong" into a failure-mode taxonomy engineering can act on, fast enough to change the current release.

Here, I help teams stand up the same rigor in days, not quarters:

• Evaluation frameworks - scoring dimensions, rating scales, accept/reject criteria, and golden sets tailored to your product, not generic checklists.

• Annotation operations - guidelines with worked examples real annotators can follow consistently, Label Studio configurations ready to import, and inter-annotator agreement analysis so you know your labels are trustworthy.

• Conversation quality audits - I read your AI agent's real transcripts, build a badcase taxonomy, cluster root causes, and hand you a prioritized fix list instead of a vague "quality score".

• Intent taxonomies - multi-level classification schemas with decision rules precise enough for both human annotators and LLM classifiers.

• Bilingual delivery - native Mandarin Chinese, professional working English. I evaluate and annotate in both languages, including mixed-language data.

How I work: async-first and documentation-heavy. Fixed-price milestones with clearly defined deliverables. You get artifacts your team can keep using - rubrics, guidelines, configs, reports - not just my opinion.

If your team is shipping an AI feature and the honest answer to "how do we know it's good?" is "we eyeball it", that's exactly the gap I fill.

Steps for completing your project

After purchasing the project, send requirements so Zengyu can start the project.

Delivery time starts when Zengyu receives requirements from you.

Zengyu works on your project following the steps below.

Revisions may occur after the delivery date.

Rubric alignment

We agree the scoring dimensions and what counts as a failure before I read a single transcript.

Audit pass

I score every conversation, classify each failure into the badcase taxonomy, and cluster root causes (per tier).

Review the work, release payment, and leave feedback to Zengyu.