Website:
elimai.ai
Job details:
Elimai.ai is an AI-native software company building next-generation intelligent platforms for
manufacturing, healthcare, and enterprise operations. We specialize in building intelligent
agentic workflows. We combine cutting-edge LLMs, specialized language models (SLMs), and
custom agents to solve real operational problems for mid-market businesses in India.
Our tech stack spans RAG systems, MCP servers, full-stack ML, and production-grade
inference pipelines. We ship, iterate, and learn in public. No slow bureaucracy.
The Role
As an AI Engineer Intern, you'll contribute to building production AI systems that process
millions of extracted structured/ unstructured data from unstructured documents and
orchestrate multi-step agentic workflows. This isn't a "learn Python" internship—it's hands-on
engineering on real products with real users.
What You'll Build
• Agentic workflows: Design and implement multi-turn agent loops that parse emails, extract intent, fetch context from APIs, and generate structured responses using varied LLMs
• MCP servers: Build Model Context Protocol integrations to connect AI agents with internal databases, ERPs (NetSuite, SAP), and custom tools. These become moats for our product.
• LLM/SLM fine-tuning: Fine-tune small language models for domain-specific tasks (invoice extraction, PO parsing, buyer intent classification) on labeled datasets. Measure BLEU/F1 gains vs. base models.•
RAG intelligence layers: Build retrieval pipelines over RabbitMQ emails and documentstores. Implement vector embeddings (text-embedding-3-small), vector DBs (Pinecone), and semantic search.
• Full-stack ML: Train, evaluate, and deploy models end-to-end. Own the notebook → experiment tracking (Weights & Biases)→ dataset versioning (DVC)→ inference server (FastAPI)→ monitoring pipeline.
Required Skills
• Python expertise.LLM fundamentals: Understand prompt engineering, token budgeting,
context windows, and why you use different models for different tasks.
• ML basics: Training/validation/test splits, evaluation metrics (F1, precision, recall),
overfitting, cross-validation. Can read a paper and implement it.
• SQL & databases: Write efficient queries, understand indexing, normalization.
Comfortable with PostgreSQL or similar.
• Git & DevOps basics: Version control, Docker basics, CI/CD pipelines. Understand why
we containerize inference.
Curiosity > credentials: Strong fundamentals beat fancy resume items. We care that
you can debug, think in systems, and ship.
Good-to-Have
• Experience with LangChain, LlamaIndex, or similar agent frameworks.
• Fine-tuning experience: LoRA, QLoRA, or custom training runs on HuggingFace models.
• RAG at scale: Built or debugged RAG pipelines in production.
• MCP experience: Built or integrated Model Context Protocol servers. Experience
working in VLMs is a plus.
What We Offer
• ₹10,000-14,000/month stipend: Hybrid internship in Coimbatore. Open for Initial Remote work but should move to Coimbatore if things are permanent.
• Real code, real impact: Your work goes to production. You'll see users, revenue, and feedback.
• Full-stack ownership: From idea → training → deployment → monitoring. Own your responsibilities in the project end-to-end.
• Direct mentorship: No corporate gatekeeping.
• Potential extension: Strong performers can transition to a full-time role or continued contract work post-internship.
Day-to-Day Responsibilities
• Build and ship features: Implement a component (agent, MCP server, fine-tuned model)
from spec to production in 1-2 week sprints.
• Debug production issues: Investigate why an email pipeline broke, a model's accuracy
dropped, or an API returned garbage.
• Read & understand papers: Stay current. We share research on in-context learning,
retrieval optimization, and agent architectures.
• Participate in code reviews: Get feedback, give feedback. Shipping quality code is non-
negotiable.
• Experiment & measure: Run A/B tests on prompt templates, model choices, and
retrieval strategies. Document findings.
Click on Apply to know more.