Allan Valim Ribeiro da Fonseca
Fulda, Germany · allan.valim-ribeiro-da-fonseca@informatik.hs-fulda.de · allanvalin@yahoo.com.br · github.com/allan-valin · Standard CV
Research interests: computational linguistics, speech and phonetic transcription (IPA), multilingual and low-resource language technology, evaluation of language models, historical linguistics and Latin.
Education
M.Sc. Applied Computer Science (Angewandte Informatik)04/2025 – expected 2027
Fulda University of Applied Sciences, Germany
- Master's thesis: topic to be defined, starting autumn 2026
- Selected modules: Machine Learning, Cloud Computing, Graphs and Networks, Test Oriented Development, Secure Software Development
B.Sc. Applied Computer Science (Angewandte Informatik)10/2017 – 04/2025
Fulda University of Applied Sciences, Germany
- Four-year programme (240 ECTS): a foundation year in engineering sciences, then the full Applied Computer Science curriculum
- Final grade 2.3 (German scale, 1.0 = best)
- Bachelor's thesis: Machine Learning for Particle Counting (grade 2.0), supervised by Dr. Martin Kumm, second examiner Dr. Rahul Singh (GSI). Labelled pile-up and sub-threshold peaks from a synchrotron scintillation counter and trained CNNs to detect them; tuning the window size doubled validation accuracy.
- Selected modules: Automata Theory and Formal Languages, Artificial Intelligence and Machine Learning, Algorithms and Data Structures, Distributed Systems, Embedded Networking
B.A. in Linguistics, Portuguese and Latin (Bacharel em Letras)02/2008 – 06/2014
Federal University of Paraná (UFPR), Curitiba, Brazil
- Monograph: Virgílio Marão Gramático: Tradução de uma Paródia (2013), supervised by Prof. Guilherme Gontijo Flores. Partial Latin–Portuguese translation of a 7th-century parody of Latin grammar.
- Selected modules: Linguistics I–III, Morphology, Syntax, Text Linguistics, Romance Linguistics, Indo-European Linguistics I–II, Stylistics, History of Translation; Latin (six semesters), Ancient Greek (two semesters)
- Term paper: corpus study of the synthetic vs. periphrastic future (“farei” vs. “vou fazer”) across several decades of one Brazilian magazine
Papers and manuscripts
- A. Valim Ribeiro da Fonseca. A Comparative Survey and Evaluation of Language-Agnostic Speech-to-IPA Transcription Systems. Term paper, Advanced Scientific Writing, Fulda University of Applied Sciences, 2026. Code: phonetic-eval
- A. Valim Ribeiro da Fonseca, F. Stumpf, S. Rieger. Replaying Recorded Traffic in Network Digital Twins: Storage Granularity, Window Alignment, and Fidelity. In preparation, 2026. Code: digsiviz
- A. Valin Ribeiro da Fonseca. Virgílio Marão Gramático: Tradução de uma Paródia. Undergraduate monograph (in Portuguese), Federal University of Paraná, 2013.
Research experience
Term paper: A Comparative Survey and Evaluation of Language-Agnostic Speech-to-IPA Transcription Systems2026
Advanced Scientific Writing, Fulda University of Applied Sciences
- Compared seven speech-to-IPA systems (Allosaurus, Wav2Vec2Phoneme, MultIPA, Allophant, ZIPA, WhIPA/STIPA, POWSM) on VoxAngeles (95 languages, 5,416 recordings)
- Found that reducing to broad phones at scoring time lowers phone error rate by 9–22 points while barely changing feature error rate, and that published scores are not directly comparable
Research project (DigSiViz): Replaying Recorded Traffic in Network Digital Twins: Storage Granularity, Window Alignment, and Fidelity2026
Fulda University of Applied Sciences, supervised by Prof. Dr. Sebastian Rieger
- Replaced DigSiViz's 60-second in-memory history with InfluxDB, fed by Telegraf from the existing Kafka transport, with a 12-tier down-sampling scheme so recorded network telemetry can be kept and replayed
- Measured replay fidelity across storage granularities: peak accuracy depends on window size and alignment relative to event duration; storing only the fields that are read cut ingest by 97.9%
- Stack: InfluxDB, Telegraf, Grafana, Kafka, Docker, Python/Flask; manuscript in preparation (see Papers)
Undergraduate research (Iniciação Científica): A manipulação da linguagem na Idade Média: Virgílio Maro e sua paródia de gramática (Language manipulation in the Middle Ages: Virgilius Maro and his parody of grammar)08/2010 – 06/2011
Federal University of Paraná, supervised by Prof. Dr. Rodrigo Tadeu Gonçalves (voluntary)
- Part of the research project “Comédia Latina: a tradução como reescrita do gênero”
- Reviewed the scholarship on the author and his work; the topic continued as the undergraduate monograph
- Presented the project at EVINCI 2011, UFPR's undergraduate research symposium
Other projects
Speech-controlled robot: “LLM zur Steuerung des CityBots”09/2026
Master lab course (Masterpraktikum) Autonomous Mobile Robots, Fulda University of Applied Sciences; individual project
- Built a German voice-control pipeline for a ROS 2 Nav2 robot: Whisper speech recognition, a rule-based grammar with an LLM fallback (qwen3:32b via Ollama), command validation and Piper speech output
- Cut latency by moving Whisper to the lab GPU server (about 0.1 s transcription, about 0.7 s from end of utterance to motion on the grammar path), with Vosk for wake word and fast stop detection
- Developed in Gazebo and demonstrated live on a VolksBot
Team project: “Robotergestützte 3D-Kartierungspipeline” (robot-assisted 3D mapping pipeline)2026
Fulda University of Applied Sciences (team of six)
- Integrated and tuned the autonomous-exploration part (ROS 2 Jazzy, Nav2, SLAM Toolbox, an existing frontier-exploration package) that drove VolksBot robots to scan positions for a RIEGL 3D laser scanner
- Wrote the laptop/robot launch files and tuned Nav2 and exploration parameters (footprint, inflation, velocity limits) so the robots passed through doors and out of dead ends
- Ran test drives, campus Wi-Fi coverage measurements and the literature research on 3D next-best-view planning
- Drafted the start-up guide for the four robots, later merged into the team documentation
Seminar paper: “Using LLMs for Process Mining Tasks”2026
Master seminar Algorithms of Process Mining, Fulda University of Applied Sciences (with Felix Krull)
- Co-wrote the paper and gave three presentations; my part covered in-context learning vs. LoRA fine-tuning for semantics-aware process-mining tasks
Desembarque: transcription of historical passenger manifests2026
Own project
- Transcribes handwritten Brazilian ship passenger lists (Arquivo Nacional) with open-weight vision models and shows each reading next to the page for verification; built never to assert what the page does not say. Code: ocr-gene-docs
IPA Quest2024
Game project, Unity
- Metroidvania in which the player collects the symbols of the International Phonetic Alphabet, each playing its sound. Play
Further software, game and media projects: allan-valin.github.io
Professional experience
Working student, IT Service Desk03/2022 – 03/2026
EDAG Engineering Group, Fulda
- First-level IT support: user tickets, preparing and handing over leased devices
Working student, IT08/2019 – 09/2020
OB//CC GmbH & Co. KG, Fulda, 06/2020 – 09/2020; UTH GmbH, Fulda, 08/2019 – 06/2020
- Market research on the school systems of all Latin American countries; database of all 60,000+ schools in Germany; operating system setup, website maintenance and translation
Copy editor08/2015 – 07/2017
Grupo Opus, Curitiba, 04/2017 – 07/2017; CCZ Comunicação, Curitiba, 08/2015 – 11/2016
Online editor (tech journalism)12/2008 – 07/2015
Grupo NZN, Curitiba: Baixaki, TecMundo and Baixaki Jogos (later TecMundo Jogos, now Voxel)
- Reviewed software and video games for a large Brazilian technology portal; also wrote news articles
- More than 600 bylined articles on TecMundo, 2008–2015 (author page, archived 2015)
Skills and languages
| Programming | Python (research, thesis and course projects); Java, C, C++, C#, JavaScript at coursework level; basic Haskell, Dart/Flutter |
| ML and speech | TensorFlow/Keras, PyTorch (basic), NumPy, Matplotlib; speech tools (Whisper, Vosk, Allosaurus, Piper); phone-recognition evaluation (PER, PFER) |
| Data and systems | Linux, Git, Docker, InfluxDB, Telegraf, Grafana, Containerlab, Kafka; OpenStack, Terraform, Kubernetes, AWS (lab level) |
| AI-assisted development | Apps built by directing AI coding agents (Claude): specification, testing and review (Android with Kotlin/Compose, web with Svelte) |
| Robotics, games and media | ROS 2 (Nav2, SLAM Toolbox), Gazebo; Unity, Unreal Engine 5, OpenGL (C++), Meta Quest VR, Blender, Pure Data, Arduino |
| Writing | LaTeX; 7 years of professional writing and copy editing in Portuguese (news, tutorials, reviews) |
| Languages | Portuguese (native), German (C1, DSH-2), English (C1, IELTS and TOEFL), French (good, spoken), Latin (reading), Ancient Greek (basic) |
References
