ML Engineer

As the ML Engineer at ConveGenius, you will design, build, and ship production LLM-powered systems that power personalized learning in Indian languages and educational contexts. You own applications end-to-end: RAG and agentic pipelines, evaluation harnesses, and the services that serve them in production, with a strong bar for software engineering, not just model or prompt work. Fine-tuning experience is a plus for this role, not a requirement.

Responsibilities:

• Design, build, and ship production-grade LLM applications: RAG pipelines, agentic workflows, and tool-calling systems for education-specific use cases such as question answering, content generation, adaptive feedback, and curriculum alignment.
• Own full system architecture: data pipelines, retrieval layers, orchestration, APIs, and service infrastructure, written as maintainable, tested production code, not notebooks or scripts.
• Build rigorous LLM evaluation frameworks: task-specific benchmarks, regression testing, human eval loops, and automated quality gates to catch degradation before it ships.
• Own latency and cost at the application layer: caching strategies, request batching, model/route selection, and prompt-context optimization to keep production systems fast and affordable at scale.
• Build data pipelines for grounding and instruction data, including multilingual and Indic language sources.
• Assess new open-source and closed model releases for domain applicability, cost, and production readiness.
• Define and track system performance metrics, evaluation benchmarks, and reliability targets across all shipped features.
• Work with open-source and sovereign LLMs, integrating them into production-proven serving and orchestration frameworks.
• Where warranted, fine-tune or adapt foundation models (LoRA/QLoRA/SFT), a strong plus, not a gating requirement for this role.

Requirements:

• Strong production Python engineering: clean, tested, maintainable code, not scripts or notebooks. Comfortable owning services in production.
• Hands-on experience building and shipping RAG or agentic systems in production: retrieval design, orchestration (e. g., LangChain/LangGraph or equivalent), tool calling, and multi-step reasoning pipelines.
• Demonstrated experience building LLM evaluation systems: benchmarks, regression tests, human-eval workflows, and quality monitoring, not just anecdotal "it works.
• Experience building and operating backend systems/services: APIs, data pipelines, deployment, and monitoring in a real production environment.
• Working knowledge of AI/ML pipelines, data preparation, and model deployment.
• Experience working with open-source and sovereign LLMs, using standard, production-proven frameworks.
• Understanding of cloud platforms and infrastructure for serving ML/LLM systems at scale.

Good to Have:

• Hands-on fine-tuning experience with LoRA, QLoRA, SFT, DPO, or similar preference-optimization techniques.
• Familiarity with DeepSpeed, FSDP, or large-scale distributed training frameworks.
• Experience with inference optimization (quantization, ONNX, efficient serving) at the model layer.