AI Engineer Case Study

ELMO Project

Revolutionizing News Consumption Through Advanced AI-Powered Content Generation & Intelligent Aggregation

Executive Summary

As the AI Engineer for ELMO, I architected and implemented an intelligent news aggregation system that fundamentally transformed information consumption patterns. By seamlessly integrating cutting-edge language models with real-time data retrieval mechanisms, I developed a sophisticated AI pipeline capable of generating contextually relevant, non-redundant news content at scale.

Role
AI Engineer
Duration
Jan - May 2025
Team
5 members
Institution
UT Dallas

Performance

32% faster than target response time

Accuracy

94% factual accuracy in generated content

User Impact

78% improvement in comprehension scores

Efficiency

23 minutes saved per user session

Performance Metrics

Quantifiable results demonstrating the impact and effectiveness of the AI engineering solutions

85%

Content Redundancy Reduction

Target: 70%

94%

Factual Accuracy Rate

Target: 90%

3.4s

Response Time

Target: 5s

98%

English Consistency

Target: 95%

Technical Implementation

AI Architecture

Integrated DeepSeek R1:14b with custom deployment and enhancement pipeline

┌─────────────────┐    ┌──────────────────┐    ┌─────────────────┐
│   DeepSeek R1   │    │   Ollama API     │    │  Custom RAG     │
│   (14B params)  │◄──►│   (Local Host)   │◄──►│   Framework     │
└─────────────────┘    └──────────────────┘    └─────────────────┘
         ▲                        ▲                       ▲
         │                        │                       │
         ▼                        ▼                       ▼
┌─────────────────┐    ┌──────────────────┐    ┌─────────────────┐
│  Google Search  │    │   NewsAPI        │    │   Pinecone      │
│      API        │    │   Integration    │    │ Vector Store    │
└─────────────────┘    └──────────────────┘    └─────────────────┘

Key Achievements

  • Reduced hallucination rate from 23% to <5%
  • Achieved 98% English-only response rate
  • Maintained 94% factual accuracy

Performance Impact

  • 99.2% API uptime during development
  • 80% reduction in local resource requirements
  • Seamless team collaboration enabled

Custom RAG Framework

Built sophisticated Retrieval Augmented Generation combining real-time news data with AI generation

Multi-source Retrieval

NewsAPI + Google Search with intelligent filtering and deduplication

Content Deduplication

Semantic similarity analysis to eliminate redundancy across sources

Dynamic Adaptation

Adjusts output complexity based on user preferences and content type

Technical Challenges & Solutions

Key obstacles overcome during development and innovative solutions implemented

1
Model Hallucination & Language Consistency

2
Real-time Information Integration

3
Content Quality vs. Speed Trade-off

Technical Innovation

Novel approaches and frameworks developed to solve complex AI engineering challenges

Hybrid RAG Architecture

  • Semantic retrieval using vector embeddings for contextual relevance
  • Temporal retrieval prioritizing recent content and breaking news
  • Source-diverse retrieval ensuring perspective variety and bias mitigation

Source Transparency Framework

  • Comprehensive source tracking and attribution with hyperlink preservation
  • Credibility scoring and bias analysis integration
  • Fact-check history integration and verification workflows

Technologies Mastered

Cutting-edge AI and cloud technologies leveraged to build a scalable, intelligent news platform

DeepSeek R1:14b
Ollama API
RAG Architecture
Pinecone Vector DB
AWS Lambda
Prompt Engineering
NewsAPI
Google Search API

Let's Connect

Interested in discussing AI engineering, machine learning innovations, or potential collaboration opportunities? I'd love to connect and explore how we can build the future together.

LinkedIn