Daris D.

Key Skills

Software

Other

Top Subject Matter

Social network analysis and NLP using news/text corpora (New York Times)

ESG compliance fact-checking

ABSA sentiment extraction

Top Data Types

Text

Document

Image

Top Task Types

Question Answering

Entity (NER) Classification

Function Calling

Fine-tuning

Tracking

Freelancer Overview

News Popularity Research (New York Times) — End-to-end Introduction to Social Network Analysis and NLP (Graph + NLP pipe. Brings 4+ years of professional experience across complex professional workflows, research, and quality-focused execution. Core strengths include Other. Education includes Master of Science, Oulu University (2024) and Bachelor of Science, Nanyang Technological University (2020). AI-training focus includes data types such as Text, Document, and Medical and labeling workflows including Question Answering, Entity (NER) Classification, and Function Calling.

Labeling Experience

Research Engineer (Multimodal AI) - Birru Multimodal Research Implementation

ImageTracking

Worked on an end-to-end multimodal RAG chatbot and ESG report generation system for production use. Built a modular NLP/LLM pipeline to synthesize traceable ESG narratives with structured outputs and automated citations. Developed a scalable multimodal pipeline using multiple VLMs with dynamic routing, document scoring, and persistent memory. • Implemented grounded citeable responses and LaTeX report generation • Engineered text+image multimodal processing with improved contextual accuracy • Designed a four-stage LLM pipeline with prompt management and compilation • Integrated BibTeX workflows for consistent evidence tracking

2026 - Present

Birru Multimodal Research Implementation — Company Multimodal Generative Report and Multimodal Chatbot

OtherDocumentFunction Calling

Engineered an end-to-end multimodal pipeline for grounded, citeable ESG reporting and multimodal chatbot responses from structured and unstructured inputs. Implemented text+image support using multiple VLM components with dynamic routing, document scoring, and persistent memory for improved contextual accuracy. Designed modular LLM stages to support traceable narrative synthesis with automated bibliography workflows. • Built multimodal RAG chatbot using Streamlit and OpenRouter • Converted company and news data into structured LaTeX reports with automated citations/BibTeX • Implemented dynamic routing, document scoring, and persistent memory • Created a scalable 4-stage LLM pipeline for structure-to-generation and compilation

2026 - Present

Oulu Summer Traineeship and Final Year Project — Fact-Checking GraphRAG on Company ESG Compliance (multi-source ESG document processing)

OtherDocumentEntity (NER) Classification

Built an end-to-end ABSA system to extract sentiment from multi-source documents for company ESG compliance fact-checking. Processed large collections of sustainability reports, compliance materials, news, and social media content to enable scalable ESG analysis and nuance detection such as greenwashing indicators. Integrated retrieval approaches to support grounded compliance evaluation across sources. • Developed production-ready sentiment extraction pipeline (ABSA) • Built data pipelines for 80 annual reports and 10,000 news articles • Applied prompt tuning and ClimateBERT benchmarking for simple/complex tasks • Integrated RAG/GraphRAG to improve retrieval accuracy for ESG compliance

2025 - Present

News Popularity Research (New York Times) — End-to-end Introduction to Social Network Analysis and NLP (Graph + NLP pipeline for structured analysis)

OtherTextQuestion Answering

Developed an end-to-end graph and NLP pipeline to perform social network analysis tasks including automated entity extraction and metric computation. Produced research-grade analyses using centrality metrics, clustering, and community detection to explore relationships and token popularity. The work supports downstream question answering and retrieval-style analysis by structuring entities and their connections from unstructured text sources. • Automated entity extraction and graph construction • Computed social network metrics and ran exploratory visualizations • Performed clustering/community detection and Erdos-style simulations • Analyzed token-popularity correlations for interpretability

2025 - Present

Transfer Learning for Medical Image Classification — Deep Learning Final Project on Diabetic Detection from Eyes

OtherFine-tuning

Developed a transfer learning-based medical image classification framework for diabetic retinopathy detection under limited data conditions. Implemented feature extraction strategies to address domain gaps using advanced neural attention mechanisms and ensemble learning approaches. The project focused on learning discriminative representations suitable for evaluating image findings for classification. • Applied transfer learning for domain-specific variability in medical imaging • Built scalable and interpretable DR classification framework • Used attention mechanisms and ensemble learning for feature extraction • Achieved reported accuracy improvements (0.57–0.58)

2024 - 2024

Education

Work History