Ashish Som

Associate Consultant at Bridge Medical Consulting, adept at engineering scalable AI pipelines and optimizing data workflows. Proven expertise in Python and problem-solving, delivering actionable insights through advanced data analysis using AI and collaboration with cross-functional teams. Successfully enhanced document evaluation processes, ensuring high-quality outputs aligned with clinical research standards.

Currently exploring advanced generative AI concepts, including LLM fine-tuning, RAG models, and agentic AI systems.

1. TIAB Screening Automation System (LLM-Powered)

Tools & Tech: Python, OpenAI API, Pandas, PyPDF2, YAML, Excel, Batch Scripts.

Description: Designed and deployed an end-to-end automation pipeline to conduct TIAB (Title and Abstract) screening using LLMs. The system screens large volumes of biomedical abstracts and classifies them based on inclusion/exclusion criteria.

• Built a reusable config-driven engine: Enabled dynamic execution by just modifying YAML config files (file paths, flags, model settings).
• Developed a custom PDF Reader module to map each document to its corresponding metadata entry.
• Engineered a Prompt Builder that reads screening logic from Excel and dynamically generates structured prompts for the LLM.
• Integrated OpenAI APIs and implemented a response parser to extract and format model decisions into a structured Excel file.
• Added logging, testing hooks, and validation mechanisms for model confidence and quality control.

Outcome: Cut down manual screening effort by 70%, reduced processing errors, and scaled seamlessly for large datasets.

2. Full-Text Screening and Extraction Pipeline (Advanced AI Automation)

Tools & Tech: Python, PyPDF2, OpenAI API, Pandas, YAML, Excel

Description: Created a scalable AI pipeline to screen full-text research papers for complex criteria (e.g., clinical relevance, population criteria).

• Parsed full-text PDFs and mapped them to custom IDs to process and parse with consistency
• Engineered a dynamic prompt builder to generate structured prompts aligned with defined schemas for LLM-based screening and data extraction.
• Incorporated prompt engineering and iterative prompt fine-tuning workflows to optimize output quality across diverse research contexts.
• Enabled project-level customizations for prompt formats, schema variations, and rule-based inclusions.

Outcome: Significantly enhanced the scalability and accuracy of AI-assisted literature reviews, reducing manual workload and improving the consistency of evidence synthesis.

3. LLM-based Text Summarization Pipeline

Tools & Tech: Python, OpenAI API, PyMuPDF, Pandas, YAML

Description: Built a modular pipeline to automatically summarize full-text PDFs in the medical and scientific domain using LLMs.

• Ingested documents in bulk and converted them into clean text blocks.
• Engineered flexible, section-wise summarization prompts (e.g., Introduction, Methods, Conclusion) to increase relevance.
• Included a test run module for prompt tuning and iterative improvement of summary length, coherence, and factuality.

Final summaries were exported to Excel for business user validation.