Job Profile: Machine Learning Engineer - III
Location: Bangalore | Karnataka
Years of Experience: 6 - 8
Position Overview:
Swiggy is India's largest on-demand platform for food, groceries, and dining. We're one of the few companies where AI meets physical-world execution at scale: predicting demand across 5,000+ dark stores, optimizing millions of deliveries daily, and building agents that operate in real-time with real constraints.
Forward Deployed AI Engineers embed with teams across Swiggy to solve their hardest problems using AI. You'll work in small pods, report into the Central AI team, and deploy to where the problems are: supply chain, category ops, marketplace, support, and beyond.
Think of it as being the AI CTO for the domain you're embedded in.
What qualities are we looking for
Production Agentic Experience
. You've built and deployed agentic AI systems in production, not just notebooks or demos . Advanced context engineering, evaluation frameworks, and agent harness patterns . Experience with skills, subagents, tool integration, and orchestration patterns . You understand the gap between it works locally and it works at scale with real latency and cost constraints'AI-Native Mindset
. You apply first principles thinking to re-imagine existing workflows in an AI native way. . Claude Code, Cursor, or Codex is your primary environment. You ship faster with AI tools than without. . You think in skills, subagents, and parallel workflows . You can design an agent architecture, implement it, and iterate based on production feedbackProven AI Impact
. You've shipped at least one AI system that moved a real business metric. Not incremental, transformative. . You can show it to us: a repo, a production deployment, metrics before/after, or a recorded demo . You know what it takes to get from prototype to productionEngineering Foundations
. 2+ years shipping production systems with real traffic, real users, real constraints (exceptional freshers who are AI power users with proof of work will also be considered) . OR: Technical founder / early engineer at a startup with demonstrable AI impact . Trade-off fluency: latency vs accuracy, cost vs reliability, autonomy vs predictability . Strong in Python or TypeScript, LLMs, orchestration frameworks, system designCommunication & Influence
. You translate complex AI concepts to non-technical ops teams . You can lead a discovery session, whiteboard an architecture, and get buy-in in the same meeting . You write clear documentation that becomes the team's playbook . You're comfortable context-switching: ops workflows in the morning, architecture reviews in the afternoonHigh Ownership
. You find problems worth solving. You don't wait for specs. . You ship when the path isn't clear . You're energized by ambiguity and moving fast in complex environments
What will you get to do here
. Real problems at real scale: Systems that affect millions of orders, thousands of dark stores, and real P&L . Direct access: No layers between you and the teams you're solving for. You see the problem, you ship the fix. . Shape the platform: Your field work defines what the Central AI team builds. Direct input into roadmap and architecture decisions. . Founder-like ownership: You're essentially the CTO of AI for your embedded domain . Career leverage: FDE is the fastest path to becoming a technical leader who understands both AI and operations at scale . What You Get Real problems at real scale: Systems that affect millions of orders, thousands of dark stores, and real P&L Direct access: No layers between you and the teams you're solving for. You see the problem, you ship the fix. Shape the platform: Your field work defines what the Central AI team builds. Direct input into roadmap and architecture decisions. Founder-like ownership: You're essentially the CTO of AI for your embedded domain Career leverage: FDE is the fastest path to becoming a technical leader who understands both AI and operations at scaleSome Problems You'll Work On
. Why did availability drop 15% in a zone last week Build the prediction system that prevents it before it happens. . Ops teams spend 3 days onboarding new catalog items. Can an agent do it in 20 minutes . The Dineout discovery bot answers questions but doesn't learn from mistakes. Make it self-improving. . Every team wants to build AI products but keeps rebuilding the same infrastructure. Create the foundation they build on.What You'll Build
. Prediction agents for availability, inventory, and demand forecasting . MCP servers connecting AI to Swiggy's operational systems . Evaluation frameworks that prove AI impact in production . Self-improving agents that learn from ops feedback loops . Reusable skills and subagent libraries that other teams can deployHow You'll Operate
. 60% embedded with business teams (Instamart, Food, Dineout, Customer Support etc) . 40% Central AI: roadmap input, pattern codification, build platform features . Rotation model: 3 month deployments per domain . Join their standups, map their workflows, find the highest-leverage problems . Ship production systems that handle real traffic. Not decks, not demos. . Package what works into skills and playbooks others can use . Your field learnings directly shape what the Central AI platform builds next
Visit our tech blogs to learn more about some the challenges we deal with:
https://bytes.swiggy.com/the-swiggy-delivery-challenge-part-one-6a2abb4f82f6
https://bytes.swiggy.com/how-ai-at-swiggy-is-transforming-convenience-eae0a32055ae
https://bytes.swiggy.com/decoding-food-intelligence-at-swiggy-5011e21dbc86
We are an equal opportunity employer and all qualified applicants will receive consideration for employment without regard to race, colour, religion, sex, disability status, or any other characteristic protected by the law.
