AI Full Stack Developer

AI Full Stack Developer
Contact : [email protected] / 9940999250

Experience: 3–6 years
Employment Type: Full-time

Job Summary

We are seeking an AI Full Stack Developer to build AI systems from data preparation and model training through inference, APIs and user interfaces. The role requires hands-on work across model fine-tuning, alignment, agent development and production deployment.

Key Responsibilities

Model Training and Fine-Tuning

• Prepare training datasets through deduplication, quality filtering and data curation.
• Run continued pretraining, supervised fine-tuning and instruction-tuning pipelines.
• Apply LoRA, QLoRA, adapters or full fine-tuning based on task requirements.
• Work with distributed training using FSDP, DeepSpeed or parallel training approaches.
• Track configurations, datasets, seeds, checkpoints and evaluation results.
• Measure model quality and catastrophic forgetting after training.

Alignment and Evaluation

• Build preference datasets and reward models.
• Apply methods such as RLHF, PPO, GRPO, DPO, KTO, ORPO or reinforcement learning from verifiable rewards.
• Design reward functions and identify reward hacking.
• Incorporate safety and refusal behaviour into alignment datasets.
• Maintain benchmarks, custom evaluation datasets and regression tests.

Agents and Context Management

• Build agents that plan, call tools, verify results and recover from errors.
• Develop agent control loops, tool routing, state management, retries and stop conditions.
• Build MCP servers and clients to connect tools, data and services.
• Manage retrieval, memory, summarisation and context compaction.
• Evaluate task success, tool errors and cost per task.

Inference and Application Development

• Serve models using vLLM, TGI or llama.cpp.
• Apply quantisation and measure its impact on model quality.
• Optimise time to first token, throughput, latency and GPU utilisation.
• Build asynchronous FastAPI services with Pydantic validation.
• Implement streaming through Server-Sent Events or WebSockets.
• Add authentication, rate limits, quotas, audit logs and background workers.
• Build React or Next.js interfaces using TypeScript.
• Develop dataset viewers, annotation tools and evaluation dashboards.

Infrastructure and Operations

• Containerise services with Docker and deploy on Kubernetes or managed GPU platforms.
• Maintain CI/CD for code, models and datasets.
• Monitor traces, logs, token usage, latency, errors and model drift.
• Maintain rollback procedures for model releases.

Required Skills

• Strong Python skills and working knowledge of TypeScript or JavaScript.
• Hands-on experience with PyTorch, Hugging Face Transformers, TRL, PEFT, Accelerate and Datasets.
• Experience with SFT, LoRA/QLoRA, distributed training and dataset curation.
• Practical experience in at least one alignment method, with measured results.
• Knowledge of tool calling, MCP, agent loops and context management.
• Experience with model serving, quantisation and latency optimisation.
• FastAPI, asynchronous Python, Pydantic, PostgreSQL and Redis.
• Embeddings, vector databases and RAG pipelines.
• Docker, Git, Linux, CI/CD and GPU memory planning.
• Strong evaluation, debugging and technical documentation skills.

Preferred Skills

• Federated learning or privacy-preserving training.
• Custom CUDA or Triton kernels.
• Model distillation and merging.
• Guardrails, red teaming and jailbreak testing.
• Multimodal models involving vision, speech or documents.
• On-premises or air-gapped deployments.
• Relevant open-source contributions.

Application Requirements

Share an updated CV along with one of the following:

• A model training or fine-tuning project with evaluation results.
• A repository containing an agent, MCP server or inference service.
• A short case study explaining an AI problem solved and how results were measured.

Pay: ₹682,000.00 - ₹1,150,000.00 per year

Work Location: In person