Looking for a job

ay User

@ayuser

Message

I build reliable LLM workflows and retrieval systems for pharmaceutical and research products.

India

Message

At GSK, I build LLM workflows for pharmaceutical regulatory-change assessment across eight countries, covering OCR ingestion, evidence retrieval, approval classification, required-document determination, and report generation.

I redesigned a broad LLM task into a four-pass workflow with structured outputs, self-consistency voting, semantic matching, and deterministic policy validation. I also built end-to-end retrieval and achieved 85%+ SME agreement on approval type and required documentation.

Previously, I helped reduce GSK document-generation time by 70%, built FastAPI integrations that cut OCR calls by 50%, and developed scientific-research Q&A capabilities. I’ve also built deep-learning medical-imaging models, published an IEEE TENCON paper, and shipped Python, FastAPI, and MCP projects.

Experience

Work history, roles, and key accomplishments

Current

AI/ML Engineer

Current

GSK

Jun 2025 - Present (1 year 3 months)

Primary engineer on an LLM workflow for pharmaceutical regulatory-change assessment across 8 countries, spanning OCR ingestion, evidence retrieval, approval classification, required-document determination, and report generation. Redesigned a broad LLM task into a four-pass workflow with structured intermediate outputs, parallel self-consistency voting, semantic matching, and deterministic policy-r

LLM workflows OCR Retrieval Classification Semantic Matching Policy Validation Chunking Embeddings Vector Indexing Hybrid Search Evaluation

Technical Associate

GSK

Aug 2023 - May 2025 (1 year 9 months)

Contributed to CGS, a pharmaceutical document-generation system, reducing total document-generation time by 70%. Built and deployed FastAPI integrations using Docker and Azure CI/CD, and developed context-compression and private-document Q&A capabilities for JulesOS.

fastAPI Docker Azure CI CD OCR Caching Document Generation Context Compression Q&A React Agent State Management Streaming Responses Vector Indexing Telemetry

Deep Learning Intern

Infyuva

Jan 2023 - Jul 2023 (6 months)

Built fetal head-circumference measurement from ultrasound using U-Net variants and transfer learning, achieving a mean absolute difference of 1.95 ± 1.86 mm. First author of a paper accepted at IEEE TENCON.

Deep Learning U Net Transfer Learning Keras Tuner Bayesian Optimization Hyperband Medical Imaging

PRISM Developer

Samsung R&D Institute India-Bangalore

Jan 2022 - Sep 2022 (8 months)

Applied a variant of the art gallery problem to optimize camera placement around stages under coverage and physical constraints. Received recognition awarded to the top 5% of teams.

Algorithm Design Optimization Camera Placement Art Gallery Problem

Perception Subsystem Member

Formula Manipal

Feb 2020 - Mar 2021 (1 year 1 month)

Built a navigation perception pipeline using YOLOv3, PnP, SIFT, and triangulation, and trained a ResNet keypoint-regression model for track positioning. Contributed to a facial-feature-based driver-attentiveness system that received the Ather Energy Software Award.

YOLOv3 PNP SIFT Triangulation ResNet Keypoint Regression Computer Vision Driver Monitoring