Headline

Artificial Intelligence Engineer at Alumnx AI Labs

About

I'm an AI/GenAI Engineer who works on RAG pipelines, multi-agent LLM systems, and vision-language models, with a bias toward the architecture decisions that make products practical, not just impressive. I care less about "use a bigger model" and more about retrieval, structured logic, tool-calling, and using LLMs only where they genuinely earn their place. A few things I've built: a multi-agent LLM pipeline that cut manual document analysis time by 70%+, an Agentic PRD Generator deployed on real UAV documentation, and a fully offline RAG system with LangChain and FAISS. My team finished in the top 1.2% of 57,000+ teams at the Google Agentic AI Hackathon 2025 with Ekatra, an AI microlearning platform. I've also worked hands-on with MCP servers, LLM evaluation (perplexity, LLM-as-a-Judge), and vision models like SAM, DINO and YOLO on satellite imagery. Right now I'm focused on applied LLM engineering turning messy, real-world problems into something clean, useful, and technically honest. I'm happiest at the point where retrieval, agents, and product decisions meet, and I ship side projects in public so the work speaks for itself. If you're building with RAG, agents, or applied LLMs or hiring for one say hi. I'm always up for a conversation about AI, product ideas, and interesting engineering problems. 📩 Reach me at shebafernandes5@gmail.com or through my projects on github.com/ShebaFernandes

Location

Bengaluru

AI stack

claude code

gemini

github copilot

hugging face

langchain

langgraph

Work experience

Artificial Intelligence Engineer at Alumnx AI Labs, Aug 2026 - Sep 2026 · 2 mos, Hyderabad, Telangana, India

Intern at NewSpace Research and Technologies, Jan 2026 - Jun 2026 · 6 mos, Bangalore Urban, Karnataka, India, Built a 3-stage multi-agent LLM pipeline using Claude API and OpenRouter to automate PRD generation and engineering impact analysis across a 15-module UAV software platform, reducing manual analysis time by over 70%. Designed a modular AI agent architecture supporting offline execution, structured tool integration, and retry handling, enabling production-grade workflow orchestration with zero external API dependency. Developed offline RAG pipelines using LangChain and FAISS over 10,000+ document chunks with semantic retrieval and domain-specific fine-tuning, improving answer relevance by 40%. Built computer vision workflows using SAM, DINO, and YOLO for satellite imagery analysis, achieving 85%+ object detection accuracy, and integrated multimodal AI services via REST APIs.

Education

Bachelor of Engineering - BE, Computer Science, St Joseph Engineering College, Mangaluru

12th PCMB, St. Aloysius (Deemed to be University)

Public links

github: https://github.com/ShebaFernandes