Forward Deployed Engineer — AI/ML (5+ yrs MUST || After College)

MarkAI is an AI-native services company. We place forward-deployed engineers inside client teams to build and ship production AI systems. We're already live, with engineers deployed at well-known organisations. The team is hand-picked from top-tier college networks.

We're not a body shop. You'll work next to the client's engineers, own outcomes end to end, and take a system from first prototype to production.

What you'll do

Embed with client teams and turn business problems into working AI systems

Build agentic workflows and multi-step LLM pipelines with LangGraph

Design RAG systems: chunking, embeddings, vector search, reranking and evals

Build and tune classical ML models (XGBoost, LightGBM) for scoring, risk, ranking and forecasting, and combine them with LLMs where it makes sense

Fine-tune models (LoRA/PEFT, RL where needed) on proprietary data

Improve inference latency and cost (vLLM, batching, quantisation)

Ship production-grade services with monitoring, evals and guardrails

Own reliability once it's live. If it breaks, you're on it.

Must have

5+ years of software/ML engineering, with at least 2 years shipping ML or LLM systems to production

Strong Python and backend fundamentals (FastAPI, async, testing)

XGBoost / LightGBM: feature engineering, tuning, evaluation

LangGraph (or LangChain / LlamaIndex) for agents and tool-calling

Vector DBs: pgvector, Pinecone, Qdrant, Weaviate or Milvus

Embeddings: model selection, hybrid search (BM25 + dense), rerankers

LLM evals & observability: Ragas, LangSmith, Langfuse

Cloud & deployment: AWS/GCP, Docker, Kubernetes, CI/CD

Comfortable talking directly to client stakeholders, from engineers to leadership

Good to have

Fine-tuning (LoRA/QLoRA, DPO/RL), Hugging Face ecosystem

Inference serving: vLLM, TGI, Triton

Data pipelines: Spark, Airflow, Kafka

Fintech, payments or high-scale consumer background

Open-source contributions

Why MarkAI

₹25L – ₹50L, based on depth and what you've shipped

Real production problems at scale, not demos

Small team, high ownership, direct access to the founder

Exposure across multiple companies and domains