About the role
Role Requirements:
LLM Fundamentals & PromptingClear understanding of how LLMs work, including tokenization, context windows, temperature, structured output formatting, LLM guardrails, and hallucination prevention.
- Proficiency in prompting techniques such as few-shot prompting, system prompts, structured outputs, role prompting, and reasoning-oriented prompts.
- Familiarity with recent model releases, capability shifts, and architectural developments.
- Ability to run inference on local LLMs using tools like Ollama, VLLM, or Hugging Face Transformers
AI Systems & Engineering
- Moderate to strong understanding of RAG architecture, including chunking, embeddings, retrieval, reranking, and generation.
- Working knowledge of AI agents, tool use, and agent orchestration frameworks such as LangChain, LlamaIndex, AutoGen, or custom frameworks.
- Understanding of MCP (Model Context Protocol) and AI skills/tool design
- Transformer architecture basics — attention mechanism, encoder/decoder, positional encoding, embeddings
- Good foundation in neural networks, including core concepts such as layers, activation functions, loss functions, backpropagation, optimization, and model evaluation.
- Strong programming skills, especially in Python, including clean, idiomatic code, proper error handling, type hints, debugging, and the ability to build efficient algorithms for practical engineering problems.
- Ability to work with structured outputs: JSON schema, Pydantic models, data validation patterns
- Flask API development: REST endpoints, request/response handling, middleware
- Fundamental understanding of API scaling, including async and sync patterns and basic load considerations.
- Expert-level Git operations: branching strategy, PR workflows, commit hygiene, rebasing/merging, resolving conflicts, and maintaining clean collaboration history.
- Ability to independently navigate an existing, non-trivial codebase using AI-assisted tools (Cursor, Claude, ChatGPT, GitHub Copilot, etc.)
- Uses AI tools to boost velocity — but can reason through code independently and does not require AI to explain every line
- Rough working knowledge of image processing concepts: preprocessing, transformations, color spaces
- Familiarity with OCR tools and their practical limitations (Tesseract, Docling, AWS Textract, etc.).
- Basic awareness of object detection concepts (bounding boxes, YOLO-style models)
- Ability to articulate technical decisions clearly in review calls and project syncs without needing repeated prompting to explain reasoning.
- Good ability to think through solution architecture and system design, including component boundaries, data flow, API design, scalability trade-offs, reliability, and maintainability.
- Strong written communication for async updates, PRs, and documentation
Working Proficiency Requirements:
Candidates do not need to be experts in these areas, but they should have enough working knowledge to understand, discuss, and contribute.
- Docker: containerizing Python services, multi-stage builds, docker-compose for local stacks
- Deep RAG expertise: Graph RAG, hybrid retrieval, vector database internals (Pinecone, Weaviate, Qdrant, pgvector)
- Custom LLM agent design: memory management, multi-step reasoning, tool routing, state machines
- Agent observability: tracing, logging agent runs, dashboards (LangSmith, Phoenix, custom)
- LLM/VLM fine-tuning: PEFT methods (LoRA, QLoRA), GRPO, instruction tuning pipelines
- Fine-tuning object detection models for practical use cases, including dataset preparation, annotation quality checks, training configuration, and performance evaluation.
- Good working ability with vision agents, including visual reasoning workflows, image/document inputs, OCR/VLM handoffs, and tool-based orchestration.
- Good understanding of audio handling in agent workflows, including transcription, audio preprocessing basics, speech-to-text integration, and routing audio-derived outputs into downstream tools.
- Good ability to handle multi-extension documents across formats such as PDFs, images, spreadsheets, presentations, text files, and structured exports in AI pipelines.
- Good understanding of converting structured data to unstructured representations and extracting structured outputs from unstructured inputs.
- Awareness of database selection based on use case, including when to use MongoDB, Neo4j, SQL databases, vector databases, or hybrid storage patterns.
- Frontend / UI basics: HTML/CSS/JS or Streamlit for internal tooling and demos
- Ability to actively contribute to product and solution architecture discussions by proposing practical design options and understanding trade-offs.
- Awareness of the latest releases, frameworks, and modern technologies, with the ability to choose current, practical solutions instead of relying on outdated approaches.