Context & Problem
Many text analytics dashboards rely heavily on external cloud APIs, incurring substantial bills and data privacy liabilities. OmniText showcases a modular, CPU-optimized NLP dashboard running lightweight transformers locally to provide semantic analytics.
Architecture
System pipeline layout representing coordination blocks and execution pathways designed for this solution.
Implementation
- Integrated Hugging Face pipeline interfaces for sentiment analysis, entity classification, and keyword extraction.
- Built a CPU-friendly summarization pipeline using quantized DistilBART models.
- Developed a Streamlit visual workspace with interactive TF-IDF text comparison maps.
Challenges & Trade-offs
CHALLENGE: Out-of-memory errors on shared CPU hosting when loading large transformer weights.
RESOLUTION: Migrated to ONNX-quantized models and implemented aggressive memory garbage collection after model outputs are fetched.
Results & Impact
Quantitative metrics and verified outcomes obtained from benchmarks.
$0.00API Calling Costs
sub-5sSummarization Speed
92.4%Classification F1
Media Gallery
Dashboard screens and coordinate visualizations of the running system.
OmniText entity classification table visual workspace