Update app.py

app.py

@@ -64,15 +64,17 @@ def display_work_experience():
     st.markdown('## Work Experience')
     st.write("""
     
-    **Turing, San Jose, CA, USA**
-    March 2025 - Present
-    - **Data Scientist**
-        - Successfully shipped a custom evaluation-benchmark dataset for Gemini 3.0, aggregating data sources across multiple
-            domains (mathematics, finance, chemistry, biology) spanning educational levels from high-school through PhD
-        - Collaborate with AI engineers, product teams, and academic researchers to align research initiatives with business objectives and deliver data-driven solutions across cross-functional teams
-        - Fine-tuned Qwen model on multi-GPU cluster using advanced architecture, boosting overall performance by 12% and achieving significant accuracy improvements across 15+ analytical categories
-        - Design and implement robust evaluation pipelines incorporating quality assessments, performance benchmarking, and bias mitigation techniques to enhance model reliability and fairness
-    
+  **Turing, San Jose, CA, USA**
+    March 2024 - Present
+    - **Data Scientist & Applied AI Engineer**
+        - Collaborated with AI engineers, product teams, researchers, and Google DeepMind team to integrate LLM evaluation systems into production workflows using PyTorch and distributed computing
+        - Engineered comprehensive evaluation benchmarks for Gemini 3.0 by analyzing reasoning loss patterns and image loss patterns in state-of-the-art Vision-Language Models (VLMs) including o3 and Gemini 2.5 Pro, developing custom datasets across multiple domains (mathematics, finance, chemistry, biology) spanning educational levels from high-school through PhD with statistical validation methods
+        - Implemented advanced LLM fine-tuning strategies for Qwen model including Parameter-Efficient Fine-Tuning (PEFT) with LoRA and 2-stage whole model training on multi-GPU clusters, achieving 12% performance improvement across 15+ categories
+        - Developed "auto hinter" system to improve LLM reasoning, guiding models towards correct answers based on question complexity, resulting in 8% performance increment on PhD-level questions
+        - Built "auto rater" system to assess responses from leading models like Gemini 2.5 Pro and o3 custom builds, scoring across four key dimensions: completeness, coherence, clarity, and correctness
+        - Applied advanced model compression techniques including quantization and distillation methods to optimize inference performance while maintaining model accuracy for production-ready LLM deployment
+        - Designed robust evaluation pipelines incorporating ROC curve analysis, performance benchmarking, bias mitigation, and RMSE validation to ensure model reliability and efficiency
+        
     **San Jose State University, San Jose, CA, USA**
     August 2024 - December 2024
     - **Teaching Assistant**