Vignesh Kumar

Vignesh Kumar

Director, AI Engineering

About

Vignesh is the Director, AI Engineering at Ford Motor Company, where he focuses on translating cutting-edge AI concepts into tangible products and integrated system features. His expertise spans a decade in data science, bridging advanced technical execution with strategic business objectives. He specialises in areas like advanced machine learning (CNNs, RNNs, Transformers), NLP (from sentiment analysis to LLM-powered applications), and building robust, scalable end-to-end MLOps pipelines on GCP. He is deeply engaged with the latest advancements in Generative AI and Explainable AI, ensuring model transparency and responsible AI practices. Beyond his role at Ford, he actively contributes to the AI community as a speaker and mentor, particularly within the Great Lakes ecosystem. Currently, he is expanding his skillset through a dual Master's program at IIT and IIM Indore, driven by a passion for shaping the future of AI through innovation and collaboration.

Can today's AI coding agents build software that's actually ready for production? 

In this live challenge, three developers each represent a different coding agent: Claude Code, Codex, and Google Antigravity 2.0. Starting from the same React + Vite application, already connected to a live streaming LLM API, they have just 15 minutes to complete the exact same task. 

The task: Build a production-ready AI chat assistant. That means far more than getting the model to generate a response. The application must stream responses in real time, render Markdown and code blocks correctly, handle loading and error states gracefully, support file uploads, and deliver an interface that feels polished enough to ship. 

Every prompt, code change, and engineering decision happens live on stage. The audience sees both the code and the running application side by side, making it easy to follow how each coding agent approaches the same problem. Once the builds are complete, the developers come together to discuss what worked, where the agents struggled, and when human judgment still made the difference. 

The question is simple: Can today's coding agents independently build production-quality software, or are they still best used as copilots? 

WHAT YOU'LL SEE 

  • Three developers solving the exact same front-end challenge with Claude Code, Codex, and Google Antigravity 2.0.  
  • A chat application evolve from a basic text box into a polished, streaming AI interface.  
  • Live engineering decisions around component architecture, state management, streaming, Markdown rendering, code formatting, error handling, and file uploads.  
  • Every prompt, code diff, and application update shown in real time.  
  • A side-by-side comparison that makes each agent's strengths, weaknesses, and working style easy to spot. 
Read More →

Traditional web extraction pipelines fail predictably during UI mutations, dynamic Client-Side Rendering (CSR), and aggressive bot mitigation. This hack session moves beyond theoretical problems to build a Multimodal Agentic Pipeline that actively solves these challenges by treating the web purely as a visual substrate.

Drawing from recent architectural breakthroughs in visual-first scraping, we will construct an autonomous, self-healing agent that bypasses the DOM entirely. Rather than writing fragile CSS selectors, we will build a system where a Vision-Language Model (VLM) acts as a Visual-Cognitive Engine. The agent will synthesize direct product URLs, capture pristine viewports using stealth automation, visually localize pricing regions, and crucially - semantically validate its own success. If the agent visually detects a CAPTCHA, a pop-up, or a layout failure, it triggers an intelligent self-healing loop to rotate sessions and retry, yielding deterministic, high-fidelity JSON at enterprise scale.

What We'll Build (The Agentic Architecture)

  • Intelligent URL Generation: Bypassing fragile multi-step DOM navigation (category trees, pagination) by programmatically synthesizing direct deep links from product metadata.
  • Stealth Visual Ingestion: Implementing a robust capture layer using SeleniumBase (UC Mode) with automated Advertisement & Cookie Management to dismiss overlays and ensure pristine screenshots.
  • Visual-Cognitive Extraction Engine: Deploying a Multimodal VLM (e.g., Gemini 2.5 Flash) to "see" the rendered page, identify the pricing Region of Interest (ROI), and differentiate actual selling prices from promotional noise.
  • Agentic Self-Healing Loop: Engineering a state-driven feedback mechanism where the LLM acts as a dynamic validator. If the VLM visually detects a blocking event (e.g., "Verify you are human" overlay), it signals the orchestration layer to bypass it or trigger randomized session rotations, stochastic delays, and automated retries.
  • Schema-Enforced Output: Utilizing Pydantic validation to force hallucination-resistant, structured JSON extraction from the VLM.
Read More →
View All Speakers

We use cookies essential for this site to function well. Please click to help us improve its usefulness with additional cookies. Learn about our use of cookies in our Privacy Policy & Cookies Policy.

Show details