Viju Sudhi

Experience

• 

  Make A Difference

  Jun 2012 - Jun 2014

  Project Head

  Headed 120 odd volunteers under the English Project of this organization, preparing teaching materials and calendars, organizing community fund raising campaigns and researching to develop the various faculties of the organization.

• 

  Tata Elxsi

  Nov 2016 - Mar 2019

  - Worked closely with the customer and key stakeholders to develop and maintain MCAL Driver code for various Renesas microcontrollers for LIN, CAN and SPI protocols.- Attended ISO 26262 Functional Safety Trainings and performed DFMEA and DFA for ASIL-B projects.

  • Senior Engineer

    Nov 2018 - Mar 2019

  • Engineer

    Nov 2016 - Oct 2018
• 

  Institut für Automation und Kommunikation

  Aug 2019 - Apr 2021

  Student Research Assistant

  Project: XIVT- Developed "ReForm": a new algorithm and tool for automated formalization of textual requirementsusing NLP and ML. The approach involves (i) a dependency rule-based (POS, NER, etc.) and (ii) a semantic role labels based (SRL) formalization of natural language industrial requirements to test cases.Project: BiDiMoVe- Familiarized with different messages in the Cooperative Intelligent Transport Systems (C-ITSs) and developed an algorithm that could predict the waiting time for buses at a bus stop and advise the speed range for the overall bus journey. Show less

• 

  AUDI AG

  Apr 2021 - Nov 2021

  Master Thesis Student

  - Developed an explainable bilingual text retrieval system in the automotive domain aimed at users performing competitor or wage analysis.- Explored different retrieval techniques aided by dense text representations and supplemented the retrieval results with comprehensible explanation features.

• 

  Fraunhofer IAIS

  May 2022 - now

  Research Engineer

  - Developed and evaluated "In-Car Conversational Question Answering system" with an automotive industrial partner. Implemented a novel answer moderation technique "Extraction Score" which moderates system answers ~87% as how domain experts moderate answers.- Developed and evaluated "Task-oriented dialog systems" for an insurance use case. Demonstrated how parameter-efficient fine tuning of LLMs including LoRA and IA3 demands only 6% of training data compared to full-weight fine-tuning.- Built Retrieval Augmented Generation systems with further fine-tuning Retrievers including DPR and Sentence Transformers Embeddings and relying upon Vector Databases.- Other research areas: Generative Information Retrieval, Tool Learning with Reinforcement Learning, Explainability of Language Models Show less