Jun 2024 – Present · 1 yr 11 mos · Remote

• Responsible for designing and maintaining the company’s information retrieval (IR) pipeline, along with downstream Generative AI (GenAI) workflows.
• Contributed to key improvements in the IR pipeline, including:
• Developing optimal document chunking strategies tailored to different data types.
• Structuring data for efficient indexing and retrieval.
• Enhancing search mechanisms to improve precision and recall.

Jul 2023 – Jun 2024 · 11 mos · Remote

• Owned the backend’s answer generation aspect, achieving 90 percent accuracy in answers with location indicators. The location feature was implemented without using ChatGPT solely, which helped in getting our first customer.
• Enhanced GPT-4’s capabilities by employing a custom RAG pipeline without using LangChain with FAISS for embedding search. This method allowed for handling long legal texts (e.g., 200-page contracts) by sending only relevant chunks to GPT-4.
• Implemented a comprehensive feature that allows users to upload files and accurately identify contract types without relying on ChatGPT. This solution surpasses basic regex matching, effectively recognizing various contract types. By avoiding ChatGPT, costs were minimized to nearly zero, and processing speed was enhanced, as everything runs locally on the CPU without requiring API calls.
• Solely responsible for developing the front end from scratch.
• Implemented a feature to highlight text in PDFs. Utilized React PDF as a base, incorporating custom algorithms for text highlighting using a custom Boyer-Moore search. Built this feature from scratch due to React PDF’s limitations, such as the inability to highlight multiple lines and lag during searches