SDE II - ML

We are looking for an AI/ML Engineer (SDE2) who can design, build, and operate production-grade AI systems, with a strong focus on LLM applications, retrieval, memory, and personalization. This role is suited to an engineer who combines strong software engineering fundamentals with hands-on experience building real-world AI applicationsnot just notebooks, model experiments, or basic LLM API integrations.

The core responsibilities for the job include the following:

Build Conversational AI Systems:

• Develop production-ready conversational AI experiences using LLMs, structured context, and user-specific information.
• Build systems that maintain conversational context, retrieve relevant past information, and generate personalized responses.
• Improve context selection, follow-up handling, and response consistency across conversations.

Build RAG, Retrieval, and Memory Systems:

• Design and improve retrieval-augmented generation (RAG) pipelines using structured data, embeddings, vector search, and graph-based retrieval where appropriate.
• Build long-term memory and user-profile systems that extract, store, update, rank, and retrieve relevant information.
• Handle conflicting information, memory freshness, relevance scoring, deduplication, and context selection.
• Improve retrieval quality, relevance, and personalization through measurable evaluation.

Develop LLM Orchestration and Optimization:

• Design multi-step AI pipelines involving query understanding, retrieval, reasoning, and response generation.
• Optimize model selection, prompt design, tool calls, and pipeline execution.
• Reduce unnecessary LLM calls while balancing quality, latency, and cost.
• Implement fallbacks, retries, timeouts, and graceful degradation for production reliability.

Build Personalization and User Intelligence:

• Develop systems that adapt to user preferences, goals, interests, historical interactions, and evolving context.
• Combine structured user profiles, conversational memory, and retrieved knowledge to improve relevance.
• Explore lightweight ML approaches for classification, ranking, recommendation, and context selection where they outperform additional LLM calls.

Evaluate and Improve AI Quality:

• Design evaluation datasets and test cases for retrieval quality, memory accuracy, personalization, and response consistency.
• Measure quality using appropriate offline evaluations and production metrics.
• Build regression tests to catch failures in prompts, retrieval, memory updates, and model behavior.
• Use evaluation results and production feedback to drive improvements.

Own Production AI Engineering:

• Build clean, maintainable Python services and APIs for AI features.
• Take features end-to-end: design, implementation, testing, deployment, monitoring, and iteration. Instrument latency, token usage, model costs, errors, and quality-related metrics.
• Debug production failures and improve scalability, security, and operational reliability.
• Collaborate with product and engineering teams to translate user needs into reliable AI capabilities.

Requirements:

• 3-5 years of experience in software engineering, AI/ML engineering, applied ML, or a closely related role.
• Strong Python programming and software engineering fundamentals, including clean code, API design, debugging, testing, and maintainable architecture.
• Hands-on experience building and deploying AI/ML applications using LLMs, generative AI, NLP, or related technologies.
• Practical understanding of RAG, retrieval, embeddings, search, or context-selection systems.
• Experience building backend services and integrating AI capabilities into real applications.
• Understanding of evaluation methods for AI systems, including how to measure quality and detect regressions.
• Experience with production deployment, Docker, and cloud infrastructure such as AWS or an equivalent platform.
• Ability to reason about system design, latency, cost, reliability, and failure handling.
• Ownership mindset: able to independently take a feature from problem definition through implementation, testing, and deployment.

Good to Have:

• Experience with vector databases or search systems such as Qdrant, pgvector, or similar technologies.
• Experience with graph databases such as Neo4j, knowledge graphs, or graph-based retrieval.
• Familiarity with LLM orchestration frameworks such as LangChain, LlamaIndex, or comparable tools.
• Experience with Redis, PostgreSQL, S3, background jobs, queues, and asynchronous processing.
• Familiarity with Hugging Face, model serving, vLLM, fine-tuning, or model optimization.
• Experience with recommendation systems, ranking, user modeling, or personalization.
• Experience with observability and evaluation tools such as Langfuse, tracing platforms, or equivalent solutions.
• Experience with multilingual NLP, Indic languages, or domain-specific knowledge systems.
• Interest in astrology, behavioral personalization, or building AI products that improve with continued user interaction.

What Success Looks Like:

• Ship reliable AI features that are used in the MyNaksh product.
• Improve memory, retrieval, and personalization quality using measurable metrics.
• Reduce unnecessary model calls, inference costs, and response latency without compromising quality.
• Build repeatable evaluation and regression-testing workflows for AI features.
• Take ownership of production systems, identify failure modes, and continuously improve reliability and performance.