AI intern

Job Description: AI Agent & LLM Systems Intern (Fine-tuning, Evaluation, MoE, KV Caching)

Position: AI Agent Development Intern (LLM Systems)

Location: Hybrid (Office + Remote)

Duration: 6 months

Department: AI / Research Engineering

About the Role

We are seeking a highly motivated AI Agent & LLM Systems Intern with a strong interest in LLM internals, fine-tuning, model evaluation, and systems-level optimization. This work directly feeds into our upcoming platform in wealthtech space.

This role goes well beyond basic AI usage — you will work on understanding, modifying, and optimizing model behavior, including:

• Fine-tuning open-source LLMs
• Evaluating models on relevance, accuracy, ranking, and safety metrics
• Deep-dive systems topics: Mixture-of-Experts (MoE), KV caching, attention optimizations, and runtime efficiency
• Integrating optimized models into multi-agent workflows used in production

This internship is ideal for candidates who want hands-on R&D experience in LLM engineering, model mechanics, and AI agent infrastructure.

Key Responsibilities

1. LLM Fine-Tuning & Adaptation

• Assist in fine-tuning LLMs using LoRA, QLoRA, DPO, and supervised pipelines with strict GPU budgets
• Create domain-specific agents for finance, data extraction, and research workflows.
• Prepare training datasets, perform cleaning, filtering, and alignment steps.

2. Model Evaluation & Benchmarking

• Evaluate model performance using:
• Accuracy / Precision / Recall
• F1-Score
• ROUGE
• NDCG (ranking relevance)
• Domain-specific metrics
• Compare models (Q&A models, MoE architectures, small vs. large models).
• Design evaluation scripts that run on GPUs and track improvements.

3. Understanding LLM Internals (Under-the-hood Systems Work)

Hands-on exposure to topics such as:

• KV caching and inference-time acceleration
• Mixture-of-Experts (MoE) routing and efficiency
• Transformer architecture internals (attention, feed-forward, positional encodings)
• GPU memory optimization for fine-tuning
• Quantization (INT8, INT4, GPTQ, AWQ)
• Understanding tokenization, context window management, and attention scaling

4. Agent & Pipeline Integration

• Integrate tuned models into multi-agent systems using LangChain / custom orchestration.
• Create tools and APIs for agent workflows such as data extraction, reasoning, and ranking.
• Optimize inference pipelines for latency, batching, and caching.

5. Research Engineering Support

• Read and summarize research on MoE, KV caching, and training optimization.
• Experiment with model variants (SFT, DPO, RLHF-lite).
• Document experiments, lessons, and hyperparameter choices thoroughly.

Required SkillsTechnical Skills

• Strong Python skills (PyTorch experience preferred).
• Understanding of Transformers and attention.
• Basic knowledge of LLM fine-tuning (LoRA/QLoRA).
• Experience with datasets, data loaders, and training loops.
• Familiarity with GPUs, CUDA basics, or model quantization.
• Ability to analyze model outputs and compute evaluation metrics.

Preferred Skills (Bonus)

• Experience with MoE models or study of MoE architectures
• Understanding of KV cache mechanics or inference acceleration
• Experience with FAISS/Chroma vector stores
• Familiarity with multi-agent frameworks or LangChain
• Exposure to training dashboards (Weights & Biases, TensorBoard)

Soft Skills

• Strong analytical mindset and curiosity about how models work under the hood
• Ability to read research papers and convert ideas into code
• Clear documentation and structured thinking
• Comfortable working in fast, iterative development cycles

What You Will Learn

• Real-world LLM fine-tuning workflows on NVIDIA GPUs
• Implementing KV cache optimizations in agents
• Working with MoE architectures and understanding load balancing
• Benchmarking models at scale
• Deploying optimized LLMs inside production agent systems
• Building evaluation datasets across 100k+ samples
• Understanding how model internals map to performance and accuracy

Why Join Us?

• Direct mentorship in:
• LLM engineering
• World models
• Model optimization
• Work impacts real AI products in finance and knowledge automation
• Full-time Research Engineer or Agent Developer role

Job Type: Internship

Contract length: 6 months

Pay: From ₹5,000.00 per month

Application Question(s):

• Provide a summary of AI agentic framework and the topics worked, and topics which have self studied.

Experience:

• Agent development : 1 year (Required)

Work Location: Remote