AI/ML Engineer

Flatworld Solutions Pvt Ltd.; Bengaluru; On-site; INR 1,800,000 - 2,400,000; Full time

Benefits offered by this job

Learning budget

Mentorship from AI Solutions Lead

Job summary

Flatworld Solutions Pvt Ltd. is seeking a senior AI/ML engineer to design and build LLM-powered components, including RAG pipelines, document intelligence, summarisation, classification and extraction.

You will craft agentic workflows, prompts, and structured outputs, while selecting models across APIs and open-weight options with attention to latency and cost. Responsibilities include building ingestion pipelines, tuning retrieval, training ML models as appropriate, and contributing to

Qualifications

• Bachelor's or Master's in Computer Science, Data Science, AI/ML, Statistics, Engineering or equivalent practical experience.
• 4–7 years hands-on ML or software engineering experience.
• At least 2 years building LLM or Generative AI applications.
• Experience taking an AI solution from prototype to live deployment.

Responsibilities

• Design and build LLM-powered components — RAG pipelines, document intelligence, summarisation, classification, extraction, and conversational agents.
• Develop agentic workflows using tool calling, multi-step orchestration and guardrails.
• Engineer prompts, system instructions, and structured outputs; version and test them.
• Select models for tasks across commercial APIs and open-weight models with cost considerations.

Skills

Python

LLM API development

RAG & vector databases

ML fundamentals

API development

Docker

Git

Education

Bachelor's or Master's in CS/DS/AI/Engineering

Tools

Pinecone

Qdrant

Chroma

pgvector

LangChain

LangGraph

LlamaIndex

Job description

Key Responsibilities

A. LLM & Generative AI Solution Build

• Design and build LLM-powered components — RAG pipelines, document intelligence, summarisation, classification, extraction, and conversational agents — across multiple solution concepts in parallel.
• Develop agentic workflows using tool calling, multi-step orchestration, and clear guardrails and fallback behaviour.
• Engineer prompts, system instructions, and structured output schemas; version and test them like code.
• Select the right model for each task across commercial APIs (Anthropic, OpenAI, Google, Azure OpenAI, AWS Bedrock) and open-weight models, balancing quality, latency, and cost.

B. Data, Retrieval & Model Development

• Build ingestion pipelines for client data: document parsing (PDFs, scans, spreadsheets), chunking strategies, embedding generation, and metadata enrichment.
• Design and tune retrieval — vector search, hybrid (keyword + semantic) search, re-ranking, and query rewriting.
• Build, train, and evaluate classical ML models (classification, forecasting, anomaly detection) where the problem calls for them rather than an LLM.
• Assess client data readiness during discovery and flag quality, volume, or privacy gaps early.
• Fine-tune or adapt models (e.g. LoRA) only when prompting and retrieval are not enough, backed by a clear cost-benefit case.

C. Evaluation, Quality & Cost Control

• Build evaluation harnesses for every AI component: golden datasets, automated metrics, LLM-as-judge scoring, and human review loops.
• Measure and reduce hallucinations, retrieval misses, and edge-case failures before anything goes in front of a client.
• Track token usage, latency, and cost per transaction; provide running-cost inputs for solution pricing and client ROI models.
• Implement guardrails: PII redaction, prompt injection defences, content filtering, and output validation.

D. Deployment, MLOps & Collaboration

• Package AI services as clean APIs (FastAPI or equivalent) that the Full Stack Engineer can integrate without friction.
• Containerise and deploy AI services to cloud platforms (AWS, Azure, GCP); monitor quality drift, latency, and cost in live environments.
• Support the AI Solutions Lead in pre-sales — assess technical feasibility, answer model and data questions, and contribute architecture notes to proposals.
• Maintain a reusable library of retrieval modules, evaluation scripts, prompt templates, and agent patterns so each new engagement starts further along.
• Document model choices, evaluation results, and known limitations for every build.

Requirements

Mandatory Technical Requirements

• Python: Strong production-grade Python — clean, typed, tested code; async patterns; dependency and environment management. [MANDATORY]
• LLM Application Development: Hands‑on experience building on LLM APIs (Anthropic, OpenAI, Google, or Azure OpenAI) — prompt design, tool/function calling, structured outputs, streaming, and token and cost management. [MANDATORY]
• RAG & Vector Databases: At least one retrieval‑augmented system built end to end — chunking, embeddings, vector stores (Pinecone, Qdrant, Chroma, pgvector, or similar), and retrieval tuning. [MANDATORY]
• ML Fundamentals: Solid grounding in supervised learning, evaluation metrics, overfitting, and embeddings, with hands‑on use of scikit‑learn and PyTorch or TensorFlow. [MANDATORY]
• AI Evaluation: Demonstrated practice of measuring AI output quality with test sets and metrics — not just manual spot‑checks. [MANDATORY]
• API Development & Deployment: Ability to expose models as REST APIs, containerise with Docker, and deploy to a cloud platform; proficiency with Git. [MANDATORY]

Strongly Preferred

• Orchestration Frameworks: LangChain, LangGraph, LlamaIndex, or equivalent; experience with agent frameworks and the Model Context Protocol (MCP).
• Document AI: OCR and document parsing (Azure Document Intelligence, AWS Textract, Unstructured, or similar) for messy enterprise documents.
• Cloud AI Platforms: AWS Bedrock / SageMaker, Azure AI Foundry, or Google Vertex AI.
• Observability: LLM tracing and evaluation tools such as LangSmith, Langfuse, Arize, or Weights & Biases.
• Data Engineering: SQL, pandas, and building reliable batch data pipelines.

Advantageous

• Voice AI experience — speech‑to‑text, text‑to‑speech, and real‑time voice agent pipelines with telephony integration.
• Fine‑tuning and serving open‑weight models (Llama, Mistral, Qwen) with vLLM, TGI, or Ollama.
• Knowledge graphs, graph‑based retrieval, or text‑to‑SQL systems over enterprise data.
• Computer vision or multimodal model experience.
• Awareness of data protection requirements (GDPR, HIPAA, India's DPDP Act) and how they shape AI solution design.

What We Look For (Beyond the Stack)

• Evidence over enthusiasm — you trust an evaluation score more than a good‑looking demo.
• Pragmatism in model choice: you reach for the simplest approach that works, whether that is a prompt, a classifier, or a rule.
• Cost awareness — you think about what a solution costs to run at 10,000 requests a day, not just whether it works once.
• Ability to explain AI behaviour, limits, and risks in plain language to non‑technical colleagues and clients.
• A GitHub profile, Kaggle record, published work, or side projects that show what you build when nobody assigns it.

Qualifications

• Bachelor's or Master's degree in Computer Science, Data Science, AI/ML, Statistics, Engineering, or equivalent practical experience.
• 4 – 7 years of hands‑on experience in ML or software engineering, including at least 2 years building LLM or Generative AI applications.
• At least one AI solution taken from prototype to live deployment with real users.
• Prior experience in an AI product company, an IT services AI practice, a startup, or an innovation lab is a plus.

Benefits

What We Offer

• Variety — you will build across multiple industries and AI use cases rather than tuning one model forever.
• Direct line of sight from your models to a real client decision.
• Access to current commercial and open‑weight models, with freedom to pick the right tool for each problem.
• Mentorship from the AI Solutions Lead and exposure to enterprise solutioning and pre‑sales.
• Learning budget for AI/ML upskilling, conferences, and cloud certifications.
• Competitive compensation with a clear path toward Senior AI Engineer or AI Solution Architect tracks