M.Sc. Computer Science student at Blekinge Institute of Technology (BTH), graduating June 2027. I build ML and LLM systems and test them honestly: against tuned baselines, with confidence intervals, and with evidence behind every answer.
Looking for: a spring 2027 Master's thesis in applied ML, retrieval/LLM systems, or anomaly detection.
-
Drone DSS with RAG Retrieval-augmented decision support over 617 pages of EASA regulations, deployed live. Every explanation cites its source pages; 0 invented references in 12 checked cases.
-
Calibrated Defect Inspection Industrial defect detection that can answer "uncertain". Missed defects cut from 23% to 1.5%, with 25% of parts sent to human review.
-
Zero-Day Anomaly Detection Learns normal behaviour from network flows and system-call traces. 92.6% recall; 54.3% of unseen attack families caught.
-
AgentDSS (in progress) Multi-stage LLM decision pipeline with a reproducible evaluation harness.
-
Syscall IDS Benchmark (in progress) n-gram vs association rules vs LSTM across ADFA-LD and LID-DS-2021.
Python, PyTorch, scikit-learn, FastAPI, LangChain, FAISS, Java, Spring Boot, PostgreSQL, Docker
Contact: saishyamdontha@gmail.com LinkedIn: https://www.linkedin.com/in/meghashyam-sai-dontha-a2716b281/