Artificial Intelligence Engineer

Role & responsibilities:

• Design and develop voice-based AI agents that can understand and respond naturally in multiple Indian regional languages.
• Design and train an adaptive recommendation system that analyses user interactions, trends, and content relationships to continuously optimize what, how, and when content is generated or suggested.
• Optimize AI systems for cost efficiency, speed, and scalability on local or hybrid cloud setups while ensuring security.
• Work mainly with different kinds of open source models and even paid models/services wherever necessary. Understand trade-offs (e.g., retrieval vs fine-tuning, on-prem vs API)
• Drive an open-source-first strategy, selecting, fine-tuning, and deploying models locally wherever possible, while strategically using paid services only when they offer clear advantages. Understand trade-offs (e.g., retrieval vs fine-tuning, on-prem vs API)

Preferred candidate profile

• Bachelors or Masters degree in Computer Science, Data Science, Machine Learning, or a related field.
• Strong Python proficiency with experience in ML frameworks like PyTorch.
• Experience working with open-weight models like

  Llama 3, Mistral, Gemma

  , or similar.
• Skilled in

  fine-tuning (LoRA / QLoRA / PEFT)

  or using

  RAG-based customization

  to adapt models for specific domains.
• Understand

  model quantization (4-bit/8-bit)

  and compression to optimize for performance and cost.
• Experience converting and serving models locally (

  GGUF / Ollama / vLLM

  ) or via inference servers.
• Can deploy models on

  AWS, Modal, RunPod

  or locally using

  Docker

  and

  FastAPI/Flask

  .
• Familiar with

  GPU utilization optimization

  , inference streaming, and scaling small models efficiently.
• Comfortable integrating models into backends through

  REST or GraphQL APIs

  .
• Strong hands-on with the

  Hugging Face ecosystem

  transformers, datasets, accelerate, bitsandbytes, etc.
• Experienced with

  LangChain, LlamaIndex

  , and preferably

  LangGraph

  for building intelligent agents.
• Experience building or know-how of adaptive recommendation systems using reinforcement learning (Stable Baselines3 / RLlib) or hybrid/content-based recommenders (LightFM / RecBole) to influence user behavior.
• Knowledge of

  guardrails & safety mechanisms

  output validation, domain filters, and hallucination control (e.g., Guardrails.ai, LMQL, guidance).
• Familiar with core concepts:

  context windows, token limits, grounding, reinforcement learning.

• Knowledge of

  guardrails & safety mechanisms

  — output validation, domain filters, and hallucination control (e.g., Guardrails.ai, LMQL, guidance).
• Familiar with core concepts: context windows, token limits, grounding, reinforcement learning.
• Able to clean and format instruction data for fine-tuning and evaluation.
• Work with embeddings (e.g.,

  OpenAI, BGE, Instructor, text-embedding-ada-002

  ) and

  vector DBs

  like

  FAISS, Chroma, Pinecone, or Weaviate

  .
• Experience integrating

  Speech-to-Text

  systems (Whisper, Deepgram, Coqui, IndicWav2Vec, NVIDIA NeMo).
• Experience with

  Text-to-Speech

  (ElevenLabs, Azure TTS, Bark, XTTS v2, MeloTTS, Indic-TTS).
• Understanding of

  streaming and real-time response

  handling for conversational agents.
• Knowledge of

  prompt templates

  ,

  retries

  ,

  fallback logic

  , and

  caching mechanisms

  (e.g., LangChain Memory, Redis).
• Understands

  dataset scaling

  ,

  class balance

  , and

  synthetic data generation

  (using LLMs or self-play)