Akshay Sg

I build end to end AI systems that turn complex data flows into dependable, scalable products. My work spans applied machine learning, automation, and multi-agent reasoning, with an emphasis on making models and pipelines behave predictably in real world environments.

I focus on aligning strong engineering practices with practical AI decision making, ensuring systems remain adaptable and reliable as they evolve. I’m motivated by turning intricate technical problems into solutions that feel simple, intuitive, and genuinely useful.

Associate Software Engineer at Digital.ai (2025-02 – Present)

• Developed Quick Protect, an AI-powered security analysis platform that enables sales engineers to perform comprehensive vulnerability assessments on Android and iOS applications, streamlining the process from binary analysis to customer proposal generation.
• Built automated binary decompilation pipeline using Androguard (Android) and Ghidra (iOS) to reverse-engineer mobile applications into analyzable source code.
• Implemented LLM-powered code deobfuscation module that detects and reverses obfuscation patterns, making previously unreadable code accessible for security analysis.
• Designed LangGraph-based orchestration workflows to manage multi-stage analysis processes with conditional routing based on code characteristics.
• Developed AI-Quick-Insight, a business intelligence platform that enables non-technical users to query complex databases using natural language, automatically generating visualizations from conversational inputs.
• Designed semantic search to identify relevant database tables integrated with LLM-based ambiguity detection that asks clarifying questions and rephrases unclear user inputs.
• Developed natural language to SQL converter that generates optimized, validated queries from refined user questions, maintaining database security and query accuracy.
• Created end-to-end pipeline delivering query results as JSON to frontend, automatically generating appropriate charts and graphs for instant data insights.

Machine Learning Engineering Intern at Matrice.ai (2024-06 – 2025-01)

• Built a no-code computer vision platform enabling users to train, annotate, export, and deploy models without writing code.
• Engineered cloud-based data ingestion pipeline with parallel batch processing for efficient customer uploads, achieving 75% reduction in upload time by replacing sequential zip extraction with multi-threaded folder processing.
• Engineered automated training pipeline with configurable parameters (epochs, learning rate, model selection) and integrated multiple state-of-the-art computer vision architectures including YOLO, EVA, and DINO.
• Implemented model export functionality supporting multiple formats (ONNX, TensorRT) with seamless deployment pipeline using hosted inference servers.

Data Science Intern at Sports Mechanics India Pvt Ltd (2023-12 – 2024-05)

• Developed an AI-powered cricket commentary system that combines real-time event detection with automated commentary generation to enhance live match experiences and reduce manual commentary effort by 80%.
• Built real-time notification engine using Python to analyze live match data from cricket databases, detect key events using a rule-based system, and integrate with the commentary model to dynamically enrich responses.
• Fine-tuned a quantized Mistral-7B language model for generating AI cricket commentary using enriched ball-by-ball metadata with dynamic context awareness.
• Trained a YOLO model to extract bowler metrics (ball grip, delivery type, line, length, speed) from match footage, enabling performance analysis for coaching and scouting.

B.Tech in Artificial Intelligence and Data Science – Shiv Nadar University Chennai (2021-2025)