AI Engineering Intern: Applied ML & Agentic Systems

AI Engineering Intern: Applied ML & Agentic SystemsPosition Overview Company: Yodaplus Technologies Location: Mumbai, India — On-site / Hybrid Duration: 6 months, Full-time Start Date: Immediate Conversion: Pre-placement offer based on performance About the RoleYou'll work where models meet software.Half the job is classic ML/DL: training, evaluating, and shipping models behind real product features. The other half is agentic engineering: building LLM-powered agents, tools, and evaluations that hold up outside a demo.We want someone who can work across both areas and is particularly strong in at least one. What You'll Work OnApplied ML / Deep Learning● Train, fine-tune, and evaluate models across Indian languages and scripts.● Work with transliteration and spelling variants.● Build and curate datasets.● Analyse model failures and identify opportunities for improvement.● Deliver measurable gains in accuracy, latency, and reliability.Agentic & GenAI Engineering● Build agents that plan, call tools, and complete multi-step, long-horizon tasks.● Develop MCP servers that expose internal systems and capabilities to agents.● Build RAG pipelines that return structured and validated outputs.● Experiment with different agent architectures, tools, and workflows.Evaluation & Reliability● Build evaluation harnesses using:○ Golden datasets○ LLM-as-judge evaluations calibrated against human labels○ Regression test suites● Ensure every significant model or prompt change is measured before it ships.● Add tracing and monitor model cost, latency, and reliability.● Build systems that make AI behaviour measurable and debuggable.Engineering — Non-Negotiable● Write production-quality Python that is typed, tested, and maintainable.● Use Git and participate in code reviews through PRs.● Build and consume REST APIs.● Containerise applications using Docker.● Serve models behind production APIs.● Use AI coding assistants as part of your daily workflow.● Critically review and validate AI-generated code rather than blindly accepting it. Must-Have SkillsPython & Software EngineeringClean, tested code; Git; debugging; REST APIs; basic Linux and Docker.ML FundamentalsData splits and leakage, bias–variance, precision/recall/F1, and understanding why a metric can sometimes mislead.Deep LearningYou've trained at least one model in PyTorch that goes beyond a tutorial. You understand CNNs and transformers well enough to explain how and why they work.LLM Application BasicsPrompting, tool/function calling, structured outputs using JSON Schema, embeddings, and retrieval.Proof of WorkAt least one project on GitHub — a model or an agent — that you can walk us through line by line. Good-to-Have SkillsAgent FrameworksLangGraph, OpenAI Agents SDK, Claude Agent SDK, Google ADK, Microsoft Agent Framework, or Pydantic AI.Deep knowledge of one is enough. We care more about whether you understand how to build the agent loop yourself.ProtocolsMCP — built or used a server — and A2A.Computer VisionFace detection and recognition, face embeddings, liveness detection, and anti-spoofing.Indic NLPTransliteration, fuzzy and phonetic matching, and multilingual embeddings.Fine-tuning & ServingLoRA/QLoRA, quantization, ONNX, and vLLM.Evals & ObservabilityEvaluation frameworks, tracing, and prompt versioning.AI SecurityPrompt injection, PII handling, and guardrails.This is particularly important because we work with KYC and financial data. What You'll GetReal OwnershipYour work ships into products with live customers, not a sandbox.MentorshipWeekly 1:1 with a Tech Lead and code review on every PR.Rare BreadthModel-building and agent engineering on the same desk.PerksStipend, PPO opportunity, certificate, hardware, and other applicable benefits. Who We're Looking ForWe're looking for someone who is curious, hands-on, and comfortable learning by building.You don't need to know every framework listed above. We care more about strong fundamentals, evidence that you've built something yourself, and the ability to understand systems deeply enough to debug and improve them.If you've trained a model, built an agent, shipped a meaningful AI project, or gone down a technical rabbit hole because you wanted to understand how something works — we'd like to hear from you.