AI/ML Intern

Fairdeal.Market; Gurugram District; On-site; INR 250,000 - 450,000; Full time

Benefits offered by this job

None

Job summary

Fairdeal.market is seeking an ML/AI intern to tackle end-to-end problems in a lean AI team. You will frame problems, build models, evaluate honestly, and ship to production.

The role emphasizes systems thinking, rapid hypothesis testing, and disciplined measurement of model performance in production environments.

Qualifications

• Proficient in Python and ML theory, with bias/variance, regularization, and evaluation design.
• Hands-on experience with pandas, NumPy, scikit-learn, and at least one deep-learning framework.
• Familiar with LLMs, RAG, prompt engineering and grounding techniques.
• Experience with data pipelines, preprocessing, and querying (MongoDB/SQL).

Responsibilities

• Design and implement end-to-end ML systems from hypothesis to production service.
• Build and evaluate multi-agent workflows and tool-using orchestration.
• Prototype with ML models (PyTorch/TF) and manage experiments (MLflow/Weights & Biases).
• Ensure production readiness with monitoring, testing, and retraining.

Skills

Python

ML theory

Pandas/NumPy

ML frameworks

LLMs / Generative AI

Data querying

Education

CS/IT/Data Science degree

Tools

MongoDB

SQL

scikit-learn

PyTorch

TensorFlow

LangChain

MLflow

Weights & Biases

Hugging Face

Docker

FastAPI

Evidently

Job description

Fairdeal.market runs B2B quick-commerce out of dark-store warehouses across Delhi NCR, and our AI/ML systems sit directly in the operational path logistics, computer vision, and agentic tooling that run live every day. We keep the AI team deliberately lean, which means an intern here owns end-to-end problems: framing them, building the model, evaluating it honestly, and shipping it to production. We're looking for someone who thinks in systems, not notebooks comfortable moving from a rough hypothesis to a monitored, versioned service, and disciplined about measuring whether it actually works.

Core Competencies

1. Machine Learning Foundations

• Strong Python and a working command of ML theory bias/variance, regularization, evaluation design, and when not
• to use ML
• Applied experience with pandas, NumPy, scikit-learn, and at least one deep-learning framework (PyTorch / TensorFlow)
• Comfort with LLMs and generative AI: RAG architectures, prompt engineering, and grounding techniques
• Fluency with data feature pipelines, preprocessing, and querying (MongoDB / SQL)

2. Agentic Systems & Orchestration

• Designing multi-agent and tool-using workflows planning, state management, and controlled tool invocation
• Orchestration frameworks such as LangGraph or LangChain
• Wiring agents to real systems: APIs, databases, and services via MCP or equivalent integration layers
• Building and iterating on models with scikit-learn, XGBoost/LightGBM, PyTorch, or Hugging Face Transformers
• Principled feature engineering, hyperparameter search, and experiment tracking (MLflow / Weights & Biases)
• Fine-tuning and prompt/RAG optimization for LLM-driven use cases

4. Evaluation, Testing & Retraining

• Designing evaluation harnesses that reflect the real objective classification metrics for ML, and LLM/RAG evals (RAGAS, DeepEval, LangSmith) where relevant
• Unit and regression testing for ML code (pytest), with reproducible, deterministic runs
• Monitoring for data and model drift (Evidently) and standing up retraining loops on a schedule or trigger

5. Deployment & MLOps

• Serving models as production services — FastAPI, Docker, AWS (Bedrock, SageMaker, Lambda/ECS)
• CI/CD (GitHub Actions), model registries, and dataset/model versioning (MLflow, DVC)
• Production observability: logging, latency and quality monitoring, and safe rollout practices

Nice to Have

• Kubernetes or broader cloud infrastructure exposure
• A production system you've shipped and owned, side projects included

What We Expect

• Pursuing or recently completed a degree in CS, IT, Data Science, or a related field
• Self-directed: reads source and docs, debugs independently, and closes loops without hand-holding
• Intellectually honest about results reports what the data shows, not what looks good