A six-page SELL initiation (4,000p target vs 5,050p price) built on a Python valuation model using P/E, DCF and sum-of-the-parts. The repo has 11 tests and the scripts to regenerate the PDF.
- AI Developer
- Back-End Developer
- Data Scraping Specialist
I’m a London-based AI and data-minded student intern focused on building reliable, scalable extraction and analytics pipelines with Python and LLMs.
I’m a London-based AI and data-minded student intern focused on building reliable, scalable extraction and analytics pipelines with Python and LLMs. At Palate Labs, I ship production pull requests for an end-to-end system that extracts and structures restaurant menus from websites and PDFs, including robust detection for menus spread across multiple documents. I also help generate evaluation ground truth and design cost/speed observability so experiments can measure both efficiency and downstream “taste profile” impact.
Alongside my internship, I’ve built valuation and research models through equity research projects and a momentum backtester, and I’m currently leading an LLM-powered equity screener and portfolio builder for a UK/US investment initiative. My background also includes transfer pricing benchmarking work at Deloitte and operations automation at Master Marine Services, which strengthened my focus on accuracy, documentation, and practical delivery. I enjoy combining finance and engineering to turn messy real-world data into decisions that stand up to testing.
A six-page SELL initiation (4,000p target vs 5,050p price) built on a Python valuation model using P/E, DCF and sum-of-the-parts. The repo has 11 tests and the scripts to regenerate the PDF.
A portfolio backtesting engine with 19 unit tests covering lookahead bias, cost accounting and weight drift. I ran a pre-registered walk-forward study of dual momentum against 60/40 and reported that it lost out of sample (CAGR 6.7 vs 9.7 percent).
Two Chrome extensions I built and shipped: folders, search and export for AI chats, and auto refresh with page-change alerts.
ChatShelf: https://www.twine.net/signin
Pagehound: https://www.twine.net/signin
A live pipeline that estimates port waiting time from open AIS data, refreshed every 15 minutes.
Live project: https://www.twine.net/signin
Reads Statements of Facts, applies charter party clauses and flags ambiguous entries for a human. 98.1 percent event classification accuracy against hand-labelled documents.
Live project: https://www.twine.net/signin
Method and evaluation: https://www.twine.net/signin
Shipped production Python pull requests for an LLM pipeline that extracts and structures restaurant menus from websites and PDFs at scale. Built detection for menus split across multiple documents and added a fetch-all-and-merge path behind a feature flag (merged detection; merge path awaiting a paid trial). Hand-labelled and QA’d 244 menus against live restaurant sites across two batches to create ground-truth for evaluation. Designed pipeline cost and speed observability, including token-cost aggregation and dashboard panels. Ran a 200-venue experiment to measure cost and the taste-profile impact of adding drinks menus.
I ship pull requests to Palate Labs' production Python restaurant-data pipeline: menu extraction, OCR and image handling, and the tests that guard them.
Performed benchmarking and comparability analysis using Prowess and international databases for multinational clients. Prepared transfer pricing documentation aligned to OECD guidelines and Indian rules, supporting senior consultants with compliance deliverables.
Final-year top-up. Elective: Machine Learning and Data Mining.
Post a job and invite Derezar, or contact them directly. No hidden commission.