Stellin John

Orchestrated a multi-agent workflow co-pilot that triaged requests, executed tools, and integrated with Jira/Slack to reduce MTTR by 38% and auto-resolve 41% of L1 tickets; improved retrieval precision and deployed to Kubernetes with CI/CD.

Orchestrated a multi-agent workflow co-pilot that triaged requests, executed tools (SQL/HTTP), and automated L1 ticket resolution, reducing MTTR by 38% and auto-resolving 41% of L1 tickets; deployed to Kubernetes/OpenShift with CI/CD.

Built a RAG document QA and summarization API with hybrid retrieval, streaming, and citation support that reduced hallucinations by 47% and increased citation coverage to 92%, while cutting p95 latency by 34%.

Built a RAG document QA and summarization API with hybrid retrieval and citations, reducing hallucinations by 47% and raising citation coverage to 92% while cutting p95 latency by 34% and compute cost by 22%.

Designed white-box adversarial patch attacks against face liveness models, increasing FAR by 37% and hardened detectors via adversarial training to reduce EER by 21% and false accepts by 28%. Built a reproducible PyTorch evaluation harness adopted by 5+ researchers.

Architected an agentic poster generator (LLaMA-3 + Stable Diffusion) and RAG grounding over brand assets, reducing time-to-first-design by 61% and raising compliance pass rate by 23 percentage points while lowering p95 latency by 31%.

Grade: 9.33/10

Activities and societies: Relevant coursework and projects in AI/ML, computer vision, MLOps; internships and research experience in adversarial ML and generative AI.

Pursuing a Bachelor of Technology in Computer Science Engineering with strong academic performance (CGPA: 9.33/10) and coursework/projects focused on AI, ML, and systems engineering.