- Design, build and ship production AI applications end-to-end — agentic assistants, RAG systems, document intelligence, voice agents, predictive/forecasting models, and workflow-automation agents.
- Fine-tune and adapt open-weight LLMs (Qwen, Llama, Mistral, Phi, Gemma) for company-specific use cases; own the training loop, evaluation and rollout.
- Build multi-agent systems that integrate across enterprise platforms — Oracle ERP, HRMS, Google/Microsoft Calendar, email, LinkedIn, WhatsApp Business, webhooks and custom APIs.
- Own the AI stack architecture: model selection and routing, evaluation harnesses, guardrails, observability, cost and latency optimisation, security and governance.
- Establish MLOps / LLMOps practices — evals, tracing, versioning, rollback, human-in-the-loop feedback capture.

Skills





- Fine-tuning open-weight LLMs — Qwen (including larger variants such as Qwen 2.5 32B / 72B), Llama, Mistral, Phi, Gemma — using LoRA, QLoRA and PEFT.
- Preference alignment techniques — DPO, ORPO, and awareness of RLHF-style approaches.
- GPU-based training and inference — A100 / H100 / L40S; distributed training basics; quantisation (GPTQ, AWQ, GGUF); memory and throughput optimisation.
- Self-hosted inference stacks — vLLM, Ollama, llama.cpp, TGI.
- Working with Small Language Models (SLMs) for cost-productive, specialised tasks.
- Multi-model routing and abstraction using LiteLLM or equivalent.
- Managed AI platforms — Azure OpenAI, AWS Bedrock, Google Vertex AI — and clear judgment on when to build vs buy vs host.

📌 Senior Manager - Artificial Intelligence (Chennai)
🏢 Pon Pure Chemicals
📍 Chennai