HaradeepPuneti

CACI · Ashburn, VA

Computer vision and CI/CD for multi-camera border surveillance.

• Building a PyTorch computer vision pipeline pairing YOLO detection with a ResNet-50 re-identification model to classify vehicle make, model, and color and track vehicles across multiple overlapping border camera feeds.
• Enhancing CI/CD pipelines to automate testing and deployment, while prototyping a multi-agent AI system to further streamline build and review workflows.

• PyTorch
• YOLO
• ResNet-50
• CI/CD

College of Engineering, Virginia Tech · Blacksburg, VA

Benchmarking LLMs against classical NLP for scientific entity extraction.

• Built an end-to-end NLP benchmarking pipeline comparing LLM-based entity extraction against traditional and neural baselines (spaCy, GLiNER, KeyBERT) across various scientific text domains.
• Designed a reproducible evaluation framework with fuzzy matching, precision/recall scoring, and multi-replication statistical analysis to assess zero-shot LLM performance on qualitative text.
• Authoring findings for conference publication examining LLM capability boundaries in domain-specific concept identification, with a custom full-stack annotation and analysis web app.

• LLMs
• spaCy
• GLiNER
• KeyBERT

AIRCO · Brooklyn, NY

Automating diagram translation and data workflows for renewable-fuel process design.

• Developed an automated BFD and PFD translation/generation system using Python and Aspen APIs, reducing manual diagram translation time by 15% for renewable fuel process design workflows.
• Built a computer vision pipeline using OpenCV, YOLO, and Tesseract to detect and classify components across 50+ P&ID diagrams, outputting structured JSON for downstream processing.
• Designed an AI agent system automating engineering dataset management within Aspen Basic Engineering, cutting manual data entry time by 25%.

• Python
• OpenCV
• YOLO
• Tesseract
• Aspen APIs

ITVISION360 · Plano, TX · Remote

BI reporting and data pipelines on large operational datasets.

• Authored 15+ stored procedures and views in PostgreSQL to support BI reporting, processing datasets of 1M+ rows and reducing query latency by 30%.
• Built Power BI data pipelines automating previously manual reporting workflows, delivering 20+ dashboards to stakeholders and cutting weekly report prep time by 4 hours.

• PostgreSQL
• Power BI
• SQL