Riswan Basha S

Developed Tisix-text, an AI-driven quality assurance tool for a leading German publisher, reducing text-related complaints by 70%. Upgraded Tisix-Video backend workflows with Celery and Redis, increasing task throughput and boosting processing speed by 30%. Led a 3-person team to develop a German-legal RAG system with hybrid (BM25 + Dense) retrieval techniques. Deployed AI solutions into production with robust DevOps practices, achieving up to 97% accuracy in the Tisix-text and customer deliverables. Managed up to 4 simultaneous Data & AI projects in a fast-paced startup, working directly with the CTO, resulting in a 20% increase in project budgets over three quarters. Evaluated quantized models (Llama-3.1-8B, Mistral-7B-Instruct-v0.3, DeepSeek-Distilled-Llama-R1) with adapter-based (PEFT LoRA) and model-based methods, analyzing perplexity, coherence, and memory efficiency for my master thesis.

Designed and built XCGPT using LangChain, ChromaDB, and Streamlit, integrating RAG with Llama3, achieving a 40% reduction in hallucinations, as measured by manual review across 200 benchmark queries. Deployed XCGPT on Azure App Service via GitHub Actions-integrated CI/CD pipeline, automating Docker image builds to ACR and zero-downtime rollouts to App Service. Developed an NLP-powered Automotive Threat Intelligence Platform for Bosch, achieving 50% precision in identifying relevant vulnerabilities from hacking forum data using Word2Vec and advanced data cleaning with Pandas, NLTK, SpaCy, and Gensim. Created an automation tool for configuring Hardware Security Modules, slashing manual errors by 80% and saving significant operational time.

Engineered Immobilizer software in C, ensuring compliance with MISRA coding standards for robust automotive security applications. Utilized IBM DOORS (DNG) for comprehensive requirements management, facilitating clear traceability and alignment with project goals. Led configuration and management of base application software builds, streamlining CI/CD processes for efficient and reliable deployments. Implemented a custom Fuzz Testing Tool using PyQt5 and Qt Designer, significantly enhancing automated testing capabilities and software resilience.