Almas A.

Key Skills

Software

Don't disclose

Top Subject Matter

Generative AI & LLM Applications (Prompt Engineering, RLHF, Structured Output, Document Processing)

Electronic Engineering & RF/Microwave Simulation (HFSS, Microstrip Filter Design, Circuit Simulation)

Chinese Language Tasks & Technical Writing (Native-level proofreading, document evaluation, translation review)

Top Data Types

Document

Text

Top Task Types

RLHF

Freelancer Overview

LLM-Powered Academic Document Formatting System (Personal Project). Core strengths include Claude Code, DeepSeek-chat API, and LibreOffice headless. Education includes Bachelor of Engineering, Nanjing University of Posts and Telecommunications (2023). AI-training focus includes data types such as Document and labeling workflows including RLHF, Evaluation, and Rating.

Labeling Experience

Rule-audit and clarification within LLM document formatting pipeline

Document

Performed technical evaluation and rule-correction interactions within an LLM document formatting system. The pipeline audits unhandled document elements (e.g., lists, headers, hyperlinks, footnotes) and flags them for follow-up during structured transformation. • Audit layer detects elements not covered by generated transform rules. • Feedback-driven clarification improves coverage for subsequent processing. • Structured transformation rules are applied at the XML level for accurate styling. • Workflow emphasizes reliable, validated outputs rather than free-form text changes.

2025 - Present

LLM-Powered Academic Document Formatting System (Personal Project)

DocumentRLHF

Developed an end-to-end LLM pipeline for automated .docx formatting using structured outputs validated against JSON schema. The workflow includes an interactive human-in-the-loop clarify loop where the system queries the user to resolve ambiguous or unmatched document segments. • Constraint-driven prompt design with schema, fallback, and escape-rule logic. • Human-in-the-loop interaction flow for preview and rule clarification. • Structured LLM output validated with JSON-schema and safety/round limits. • Cost-optimized pipeline using DeepSeek-chat and Claude Code collaboration tooling.

2025 - Present

Education

Work History