Stuti Goel

Architected a hybrid agentic routing framework that classifies user queries into semantic search and quantitative analytics workflows, enabling accurate execution of both RAG and data-driven insights. Designed a semantic-to-numerical analytics pipeline using Sentence Transformers, cosine similarity, and Pandas, converting natural language queries into precise counts, percentages, and trend metrics from customer feedback data. Improved retrieval recall by implementing LLM-powered query expansion and schema-validated guardrails, increasing robustness of search across vector databases.

Machine Learning Engineer at Micorza (2025-05 – Present)

• Architected a hybrid agentic routing framework that classifies user queries into semantic search and quantitative analytics workflows, enabling accurate execution of both RAG and data-driven insights.
• Designed a semantic-to-numerical analytics pipeline using Sentence Transformers, cosine similarity, and Pandas, converting natural language queries into precise counts, percentages, and trend metrics from customer feedback data.
• Improved retrieval recall by implementing LLM-powered query expansion and schema-validated guardrails, increasing robustness of search across vector databases.
• Built privacy-preserving data processing workflows by developing custom PII recognizers and tokenization mechanisms for financial and corporate entities, supporting GDPR, HIPAA, and PCI-DSS compliance requirements

Artificial Intelligence Intern at Frontera Health (2024-09 – 2025-02)

• Identified speaker tagging inaccuracies in timestamped audio recordings, which affected the analysis of child-adult interactions.
• Developed heuristic functions to refine extractions and handle edge cases, and fine-tuned Whisper’s Wav2Vec model for speaker tagging using the improved dataset.
• Enhanced speaker tagging accuracy by 30% as validated by certified behavior analysts, improving the effectiveness of interaction analysis.

Research Intern at Continental automotive (2023-06 – 2024-05)

• Developed Low-Light Image Enhancement (LLIE) algorithms utilizing Python, TensorFlow, and PyTorch to address low visibility in automotive systems under challenging lighting conditions, mitigating safety risks.
• Implemented experimental evaluations on both the LOL dataset and a proprietary dataset to enhance the model’s generalizability and effectiveness
• Achieved a PSNR of 16.9 and SSIM of 0.6 with the custom model, demonstrating improvements of 10.92% and 8.20% respectively, over baseline models.

MCA in Artificial Intelligence – Bennet University (2022-09 – 2024-05)

Bachelors of Computer Applications – Kalindi College (2019-06 – 2022-05)