Company Description
Janooma is India's first agent-driven AI marketplace, enabling seamless buyer-seller interactions by leveraging an innovative conversational layer. With a database of over 4 million businesses across India, Janooma delivers live quotes from verified vendors to buyers and connects service providers with high-intent leads through WhatsApp. This platform facilitates everything from electronics to home services in one conversation, providing convenience and efficiency for all users. By solving the cold-start problem before its launch, Janooma is poised to revolutionize the marketplace with upcoming expansions to 10 Indian cities by 2026 and Southeast Asia by 2027.
LLM Systems Engineer
Location: Bengaluru (Hybrid)
Experience: 0-1 year (Freshers from 2024-2026 batches are strongly encouraged)
We are building next-generation large language model infrastructure focused on advanced model composition, optimization, and continuous capability enhancement across multiple domains and categories.
About the Role
You will play a foundational role in designing and implementing modular, scalable architectures for working with large language models. This includes building automated systems for model discovery, intelligent model merging/composition, architecture parsing, performance evaluation, gap analysis, and iterative improvement pipelines that power multiple specialized LLMs.
You will work on end-to-end LLM engineering workflows — from low-level model introspection to high-level automated training and deployment loops — enabling the creation and continuous evolution of high-performance models across various categories and use cases.
Key Responsibilities
- Design and implement automated pipelines for discovering, evaluating, and selecting high-performing open-source LLMs for different categories
- Develop advanced model composition techniques (union-style capability aggregation and intersection-style conservative merging) using state-of-the-art merging frameworks
- Parse and introspect LLM architectures in detail (similar to DOM tree parsing) — working with layers, attention mechanisms, state dictionaries, and parameter structures
- Build and maintain iterative improvement loops: task processing, multi-model comparison, automated evaluation, knowledge gap detection, synthetic data generation, and targeted fine-tuning/merging
- Implement efficient fine-tuning workflows using parameter-efficient methods (LoRA/QLoRA, PEFT) and delta adapter techniques
- Create robust evaluation frameworks using multiple metrics (ROUGE, BERTScore, faithfulness, hallucination detection, etc.) and LLM-as-Judge systems
- Optimize inference pipelines for high throughput using tools like vLLM or TGI
- Set up MLOps practices for model versioning, experiment tracking, documentation, and deployment to Hugging Face Hub
- Contribute to building reusable, modular components that can support multiple LLM categories and future domain expansions
Requirements (Freshers Welcome!)
- B.Tech / M.Tech in Computer Science, Artificial Intelligence, Machine Learning, or equivalent (2024–2026 batch)
- Strong proficiency in Python and PyTorch
- Hands-on experience with Hugging Face Transformers library
- Practical experience fine-tuning LLMs (Llama-3, Qwen2, Phi-3, or similar) using PEFT/LoRA/QLoRA
- Solid understanding of Transformer architecture (attention mechanisms, layers, embeddings, LayerNorm, MLP blocks, etc.)
- Comfort with model introspection tools (named_modules, state_dict, named_parameters)
- Good knowledge of Git, Linux, and basic MLOps workflows
- Strong problem-solving skills and excitement about working at the systems level of large language models
Big Advantages (Not Mandatory)
- Exposure to model merging frameworks (mergekit or similar) and techniques like DARE, TIES, or SLERP
- Experience with synthetic data generation or distillation methods
- Familiarity with preference optimization (DPO, ORPO, etc.)
- Experience running large models with vLLM or Text Generation Inference (TGI)
- Active Hugging Face profile with public fine-tunes or experiments
- Understanding of evaluation benchmarks and LLM-as-Judge patterns
What We Offer
- Direct ownership of critical components in a cutting-edge LLM engineering stack
- Access to high-end GPU cloud resources (A100/H100)
- Fast learning curve and rapid growth — strong performers can move into senior LLM Architect roles quickly
- Opportunity to work on multiple LLM categories and advanced model evolution systems
- Collaborative environment with clear architecture guidance and mentorship
This is a high-impact role where you will gain deep, production-grade expertise in the most advanced areas of LLM engineering in 2026.
How to Apply
Please send your resume + GitHub profile link + a brief note or link showcasing one relevant LLM project (fine-tuning, merging, evaluation pipeline, or inference work) to [Confidential Information]
Subject Line: LLM Systems Engineer – 2026
We review applications on a rolling basis.
