AI/ML Engineer (GenAI & LLM Specialist)

Role Summary

We are seeking an experienced AI/ML Engineer to lead the development of our core intelligence engine. In this role, you will own the end-to-end lifecycle of our proprietary AI tool: selecting the optimal open-source Large Language Model (LLM), fine-tuning it for our specific domain, and architecting a RAG (Retrieval-Augmented Generation) pipeline.

Your primary mission is to build a system that can ingest complex PDF documents, comprehend historical and referential data, and provide accurate, context-aware answers to user queries.

Role & responsibilities

1. Model Selection & Strategy

• Evaluate Open-Source Models:

  Analyze and benchmark state-of-the-art open-source models (e.g., Llama 3, Mistral, Falcon, Mixtral) to identify the best balance of performance, inference cost, and license suitability for our specific use case.

• Feasibility Analysis:

  Determine when to use Retrieval-Augmented Generation (RAG) versus Fine-Tuning (or a hybrid approach) to ensure the highest accuracy in answering queries based on uploaded PDFs.

2. RAG Pipeline & Data Engineering

• Document Ingestion:

  Build robust pipelines to parse, clean, and OCR complex PDF files (handling tables, headers, and multi-column layouts) using tools like Unstructured, PyMuPDF, or LayoutLM.

• Vector Database Management:

  Design and implement vector search architectures (using Pinecone, Milvus, ChromaDB, or Weaviate) to store and retrieve high-dimensional embeddings efficiently.

• Context Optimization:

  Optimize "chunking" strategies and context window management to ensure the LLM receives the most relevant historical data without hallucinating.

3. Model Training & Fine-Tuning

• Fine-Tuning:

  Implement efficient fine-tuning techniques (PEFT, LoRA, QLoRA) on the selected LLM to adapt its tone and reasoning capabilities to our business domain.

• Dataset Preparation:

  Curate and format training datasets from internal data to improve the model's ability to understand domain-specific terminology found in the PDFs.

4. Deployment & Optimization

• Inference Optimization:

  Optimize model latency and throughput using quantization (GGML/GGUF/AWQ) or engines like vLLM and TGI.

• API Development:

  Wrap the AI engine in a robust API (FastAPI/Flask) for integration with our front-end application.

Required Skills & Qualifications

• Core Tech:

  Expert proficiency in

  Python

  and deep learning frameworks (

  PyTorch

  or TensorFlow).

• LLM Ecosystem:

  Deep familiarity with the

  Hugging Face

  ecosystem (Transformers, Accelerate, PEFT, Datasets).

• GenAI Frameworks:

  Hands-on experience with orchestration frameworks like

  LangChain

  or

  LlamaIndex

  specifically for building RAG applications.

• Vector Search:

  Experience working with Vector Databases (Pinecone, ChromaDB, Elasticsearch, or pgvector).

• Document Processing:

  Experience extracting clean text from unstructured files (PDFs) using open source OCR tools (Nougat, Surya, PaddleOCR) or Python libraries(pymupdf, pdfplumber) for native pdfs.

• Deployments:

  Experience with

  Containerization & Orchestration

  (Docker, Kubernetes) and serving LLMs in air-gapped or offline environments using tools like

  vLLM, Ollama, or llama.cpp

• Mathematics:

  Solid understanding of linear algebra, probability, and how Transformer architectures (Attention mechanisms) work.

Nice-to-Have (Bonus Points)

• Experience deploying LLMs on cloud GPUs (AWS SageMaker or RunPod).
• Knowledge of prompt engineering techniques (Chain-of-Thought, ReAct).
• Previous experience building "Chat with your Data" style applications.

Why Join Us?

• High Impact:

  You will be the primary architect of the intelligence behind our product, not just maintaining legacy code.

• Cutting Edge:

  You will work with the absolute latest developments in the Open Source LLM space.

• Autonomy:

  You will have the freedom to choose the tech stack (Models, DBs, Frameworks) that best solves the problem.