ML · LLM · AI Automation Engineer
Building production-grade AI systems — from fine-tuned LLMs and agentic RAG pipelines to full-stack automation platforms that ship real business impact.
I'm an MS Computer Science graduate from the University of Bridgeport (May 2026), specializing in building and deploying end-to-end AI systems that solve real problems.
My experience spans the full AI engineering stack — from fine-tuning LLaMA 3 with QLoRA and building stateful LangGraph agents to shipping a full-stack roasting automation platform at SoundCoffees that eliminated 100% of raw material waste.
I work across both open-source LLMs (LLaMA, Phi-3) and closed APIs (Claude, GPT-4o, Gemini), with hands-on experience in MLOps, guardrails, hybrid-search RAG, and cloud deployment. My cybersecurity background gives me a unique edge in building AI systems that are both powerful and safe.
Production-grade agentic RAG system supporting 4 LLM backends (LLaMA 3, GPT-4o, Claude 3, Gemini). Features hybrid search with BM25 + vector retrieval, cross-encoder reranking, layered guardrails with PII masking and toxicity filtering, and stateful LangGraph agents with persistent memory.
Full-stack AI operations platform built from scratch during internship. Integrated live sales data via REST API, ML demand forecasting, real-time inventory monitoring, threshold-based alert systems, and a React + Supabase production dashboard used daily by operations staff.
Production AI agent built on Google Vertex AI Agent Builder using Gemini. Implemented multi-step agentic workflows with LangGraph, MCP for external tool integration, Qdrant for self-hosted vector retrieval with metadata filtering, and eval-gated deployments ensuring production reliability.
Custom training framework applying reward signals directly through transformer attention blocks — using a -10 penalty gradient on incorrect predictions to reshape attention patterns. Bridges supervised learning and RLHF principles. Tracked with Weights & Biases, visualized via Gradio.
Q-learning agent trained using experience replay buffer and custom reward shaping (+1 catch / -1 miss). Tuned epsilon-greedy exploration decay and replay buffer hyperparameters. Achieved a 96% win rate across 100 fully autonomous evaluation games.
I'm actively looking for AI Engineer, ML Engineer, and AI Automation roles starting May 2026. Open to full-time positions, contract work, and interesting conversations.
ganeshreddy1811@gmail.com