Research into real systems.

Founding engineer on an agentic sales-automation platform that turns complex RFQs into accurate quotations. I work across AI workflows, backend architecture, data models, cloud infrastructure, enterprise integrations, and product delivery.

• ↗Re-architected the backend from a monolithic prototype into a production-grade, microservices-based FastAPI system and redesigned the database schema to store complex quotation data more efficiently.
• ↗Established the engineering foundation with Terraform-managed infrastructure, GitHub Actions CI/CD, and shared development standards and guidelines.
• ↗Built an agentic pipeline that parses DXF/PDF engineering drawings, classifies requirements with LLMs, and auto-fills Excel quotations—cutting turnaround from weeks to under an hour.
• ↗Integrated inbound and outbound platform workflows with Outlook and Odoo ERP.

Built production data and applied-AI systems for healthcare clients, spanning retrieval-augmented generation, large-scale inference, web data acquisition, and entity extraction from unstructured records.

• ↗Developed a production RAG pipeline that used LLMs to generate query filters, summarize retrieved healthcare records, and rank results by similarity, with parallel processing to reduce execution time.
• ↗Designed web crawlers with Selenium, Requests, BeautifulSoup, asyncio, and aiohttp, improving pipeline efficiency by 85%.
• ↗Extracted key entities from unstructured data through classification and LLM prompt tuning, filling gaps in master datasets and improving downstream analysis.

Built Docketry, an intelligent document-processing product, contributing across applied ML, backend architecture, secure deployment, integrations, and product analytics.

• ↗Fine-tuned LayoutLMv2 to a 94% F1-score across 10 document categories, enabling accurate automated document classification.
• ↗Architected Docketry’s backend using Django and a gRPC service framework with authentication and authorization, deployed on Azure behind Nginx.
• ↗Published the Docketry PyPI package to provide plug-and-play integrations for client systems.
• ↗Applied prompt engineering to extract structured fields from raw documents and improve automation workflows.
• ↗Designed analytics dashboards that turned product usage logs into clear operational summaries.

Conducted NLP research with Prof. Pushpak Bhattacharyya, focusing on multilingual and code-mixed language understanding, low-resource data augmentation, and neural text generation.

• ↗Engineered a named-entity recognition model for code-mixed queries across 5 Indian languages, 17 entity tags, and more than 100,000 annotated entries.
• ↗Fine-tuned an ensemble BERT model to a 94.79% F1-score and developed data augmentation methods that improved targeted low-resource language performance by 20%.
• ↗Explored RNN, LSTM, Transformer, T5, and GPT architectures across NER, transliteration, machine translation, and summarization.