AI Architect (GenAI, SLM & AI Practice) (Gurugram)

AI Architect GenAI, SLM & AI Practice

Position Overview

We are looking for a hands-on AI Architect who can lead and build our AI/GenAI practice, with a strong focus on designing and developing domain-specific Small Language Models (SLMs) and enterprise AI solutions.

The ideal candidate will have deep practical experience across LLMs, SLMs, RAG, AI Agents, model training/fine-tuning, data engineering and cloud AI platforms. The person should be capable of taking an industry-specific business problem, determining the appropriate AI/model approach, defining the required components, building the model/solution, and taking it into production.

This role requires someone who can architect as well as engineernot just consume existing foundation models or build applications on top of APIs.

Key Responsibilities

• Lead AI & GenAI Practice
• Define and drive the organization's AI/GenAI technology strategy and practice.
• Establish AI architecture standards, reusable frameworks, accelerators and reference architectures.
• Identify high-value AI opportunities across different industries and customer environments.
• Lead development of AI solutions from ideation PoC model development production deployment.
• Build and mentor AI/ML/GenAI engineering teams.
• Provide technical leadership for AI solutioning, proposals and customer engagements.
• Establish best practices for AI governance, security, evaluation, responsible AI and cost optimization.
• Domain-Specific SLM Architecture & Development

A key responsibility of this role is to design and build Small Language Models for specific industries/domains.

The candidate should be able to independently determine and define the key components required to build a domain-specific SLM, including:

• Define the business and industry problem that the SLM needs to solve.
• Determine whether an SLM, LLM, RAG or a combination of these is the appropriate approach.
• Define the required model architecture and model size based on the use case.
• Evaluate whether to:
• Build a model from scratch
• Adapt an existing open-source foundation model
• Fine-tune an existing LLM/SLM
• Use knowledge distillation
• Use LoRA/PEFT or other parameter-efficient techniques
• Combine an SLM with RAG and external tools
• Define the required industry/domain-specific datasets.
• Establish data acquisition, cleaning, normalization, deduplication and preparation processes.
• Define data labeling and annotation requirements.
• Establish the required tokenizer, vocabulary and domain terminology strategy where applicable.
• Define training, validation and test datasets.
• Select appropriate model architecture, parameters and training methodology.
• Define training infrastructure, GPU requirements and distributed training requirements.
• Establish model evaluation benchmarks specific to the industry/use case.
• Define accuracy, hallucination, latency, throughput and cost targets.
• Design model compression, quantization and optimization strategies.
• Design inference architecture for cloud, private cloud or edge deployment.
• Establish model monitoring, evaluation and continuous improvement mechanisms.
• SLM Design & Engineering

The candidate should have hands-on experience with several aspects of the SLM lifecycle:

Data Model Training Fine-Tuning Evaluation Optimization Deployment Monitoring

Responsibilities include

• Design domain-specific datasets for model training and fine-tuning.
• Develop data pipelines for large-scale training data preparation.
• Select and evaluate suitable open-source/base models.
• Fine-tune models for domain-specific terminology, reasoning and tasks.
• Implement parameter-efficient fine-tuning techniques such as LoRA/QLoRA/PEFT.
• Explore knowledge distillation from larger models into smaller models.
• Apply quantization and model compression to reduce inference cost.
• Optimize models for latency, memory consumption and throughput.
• Design efficient inference architectures.
• Build evaluation frameworks to compare models and approaches.
• Establish automated model testing and regression evaluation.
• Implement continuous model improvement based on production feedback.
• LLM, RAG & Agentic AI

In addition to SLM development, the candidate should have strong hands-on experience with:

• Large Language Models (LLMs)
• RAG architectures
• Embeddings and vector databases
• Semantic and hybrid search
• Prompt and context engineering
• AI Agents and Agentic workflows
• Tool/function calling
• Multimodal AI
• Model routing and model selection
• LLM evaluation
• Guardrails and hallucination mitigation

The architect should be capable of determining the optimal architecture, for example:

SLM + RAG + Enterprise Data + AI Agents + Business APIs

rather than defaulting to a large commercial LLM for every use case.

• Industry-Specific AI Solutions
• Work with domain experts to understand industry-specific terminology, workflows and business processes.
• Convert industry .