Ari Nubar BoyacıoğluCurriculum vitae

Data Engineer

ValueWorks GmbH · Karlsruhe · Jan 2025 – present

• Own end-to-end data pipelines, from collection through transformation to KPI calculation, on a multi-tenant financial planning and reporting platform used by CFOs, boards and investors. 55+ integrated source systems spanning accounting and ERP, CRM, billing and subscription management, HR and banking, including Salesforce, HubSpot, Personio, Stripe, Zuora and Microsoft Business Central. The platform serves 90 to 100 active customers; I have shipped directly for around thirty of them.
• 146 Jira tickets delivered end to end, from requirement through implementation to production, and 490 merged pull requests between February 2025 and July 2026, with work shipped in every one of those eighteen months.
• Model financial data for reporting and planning: group consolidation, intercompany elimination, adjustment levels, deferred revenue recognition, budget against actuals and cash-flow statements, translated out of group consolidation systems and accounting ledgers into customer-facing dashboards and drill-downs.
• Built the liquidity planning and expense management engine over raw accounting exports, the longest single workstream of my time there: clustering noisy free-text booking descriptions into normalised vendors, deriving contracts from the resulting groups, classifying payment cadence from transaction sequences, and projecting each contract's next expected payment to determine whether it was still running or had churned. Replaced manual revenue and expense classification and drives a rolling cash forecast.
• Designed the per-customer data products layer: curated analytics-ready tables on a Dremio/Iceberg lakehouse and a read-only API with manifest, endpoint-scoped tokens and snapshot time travel, giving customers governed access to their own transformed data.
• Cut peak memory of the platform's largest pipeline, 400+ KPIs over ~30 GB of staging data, from over 50 GB to about 13 GB, with streaming I/O, monthly partitioning and lazy evaluation in Polars, keeping it inside its Azure Container Instance budget.
• Moved from taking requirements through Customer Success to running specification and troubleshooting calls directly with customers' finance teams, largely in German, turning their reporting logic into pipeline specifications and auditing KPI discrepancies across four pipeline layers against their live systems.
• First on the team to adopt AI coding assistants and, later, agentic development tooling; built the internal tooling that made them work against our codebase and drove adoption across engineering. Agent onboarding documentation with auto-generated code and KPI indexes, a static KPI dependency graph powering an affected KPIs pull-request review hook, reusable agent skills for tracing and documenting KPIs, and autonomous Jira-ticket-to-pull-request workflows, now part of how the team works. Personal delivery roughly doubled over the same period, from around twenty merged pull requests a month to around forty.

IT and Language Consultant

Calouste Gulbenkian Foundation, Armenian Communities Department · Lisbon, remote · Mar 2025 – present

• Evaluated grant proposals in language technology, education and digital culture across the department's funding cycles, producing written assessments with fund, partial-fund or decline recommendations for the director.
• Drafted the department's multi-year strategy for Western Armenian in digital environments, including case studies of Basque, Icelandic, Welsh and Sámi language technology programmes.
• Managed funded projects end to end: award letters, instalment schedules, interim reporting and closure, across software localisation, lexicography and mobile keyboard work.
• Acted as technical counterpart between the foundation and its grantees, and argued open-licence conditions into grant terms so that funded work could be built on afterwards.

Independent machine translation work

Karlsruhe · 2024

• Produced large-scale Western Armenian translation sets with the machine translation system built for the thesis, in collaboration with a diaspora language-revitalisation project, and extended and corrected the underlying parallel corpus from the output.

Student Research Assistant

Karlsruhe Institute of Technology, AI4LT Lab, Institute IAR · Karlsruhe · Mar 2023 – Aug 2023

• Built a streaming bot that joins virtual meetings automatically and feeds audio to the institute's Lecture Translator for simultaneous translation.

Student Research Assistant

Karlsruhe Institute of Technology, Critical Information Infrastructures, Institute AIFB · Karlsruhe · Nov 2021 – Aug 2023

• Implemented and tested modules for a blockchain-based academic publication platform (Truffle, Ganache).
• Built a configurable cover-page generator and integrated it into the thesis submission pipeline, and automated recurring research tasks such as forward and backward citation search and tabular extraction.

First neural machine translation system for Western Armenian

Master's thesis, KIT · SIGUL at LREC-COLING 2024 · with Prof. Jan Niehues

• Built the first parallel corpus for the Western Armenian and English pair: about 147,000 sentence pairs collected from websites and printed books across Germany, Turkey and Armenia, of which 52,879 are released openly.
• Fine-tuned Meta's NLLB, which had seen only Eastern Armenian, and measured how far a related-variant model transfers; published held-out-domain test sets and a negative result on custom segmentation.
• Paper in the ACL Anthology; corpus on GitHub; model served publicly with user feedback collection.

Continuing work

2024 – present

• hyw-corpus, an open language infrastructure project: corpus expansion and normalisation, frequency lexicon, morphological data, spellchecking and locale data, with a documented source manifest and reproducible pipeline.
• Speech recognition for Western Armenian, fine-tuned on the little aligned audio that exists.
• Dictionary browser extension: select a word on any page and look it up in the Western Armenian dictionaries worth using. Published on the Chrome Web Store and Firefox Add-ons.

M.Sc. Computer Science

Karlsruhe Institute of Technology · Oct 2020 – Nov 2023

• Specialisation profile in artificial intelligence. Thesis on neural machine translation for Western Armenian, published at SIGUL / LREC-COLING 2024.

B.Sc. Computer Science

Karlsruhe Institute of Technology · Oct 2015 – Sep 2020

• Thesis: a longitudinal study of how smartphone usage changed in Germany and Turkey during the 2020 social distancing period.

German Abitur

Deutsche Schule Istanbul · Istanbul · 2015

• One of two schools in Turkey able to award the German Abitur. It is why I speak German and why I ended up here.

• Self-hosted LLM stack. llama.cpp and vLLM serving quantised Qwen models on a single RTX 4090, behind a LiteLLM proxy, with Open WebUI, local search over MCP and a notification relay. Used daily as working infrastructure rather than a demo.
• Home server. Ubuntu, Docker, Tailscale, Pi-hole, running continuously and hosting the projects below.
• Health tracking application. React, Flask and SQLite, built for a specific dietary protocol and actually used.
• Transit departure board. Raspberry Pi and Flask, live Karlsruhe departures for one stop.
• narinmatbaasi.com. Next.js and Tailwind site for a historic letterpress print shop in Beyoğlu, with Armenian typography as its visual identity.