Forward Deployed AI Engineer

About UsQuanXAI is an AI technology firm delivering intelligent, scalable, and secure automation solutions for enterprises. We bridge cutting-edge AI capabilities with complex business infrastructure — from the shop floor to the boardroom — with deep expertise in manufacturing and industrial environments. Our engineering DNA combines modern AI/ML (LLMs, computer vision, predictive analytics) with industrial data systems (SCADA, MES, historians, IIoT), so our solutions work in the real world, not just in notebooks.The Forward Deployed AI Engineer's goal is to turn AI into measurable business value at the client's site. While our platform team ensures systems are scalable and deployed on AWS, you will focus on the Engagement: discovering the real problem, building the POC that proves value on real data, demoing it to stakeholders from operators to executives, and delivering it into production. You will work alongside Senior AI Engineers and Solution Architects, embedded directly with enterprise clients — offices, factories, and plants.Job ResponsibilitiesAs a Forward Deployed AI Engineer, you will own client engagements end-to-end and take on the following:1. Client Engagement & Solution Delivery (35%)•   Discovery & Requirement Gathering: Run structured workshops with business and operations stakeholders; map processes, data sources, and pain points; define quantified success criteria upfront (e.g., “detect ≥95% of defect type X with <2% false positives”, “reduce unplanned downtime on Line 3 by 15%”).•   Rapid POC Engineering: Build credible working prototypes on real client data in 1–4 weeks using FastAPI + Streamlit/Gradio or custom frontends. Handle messy reality — incomplete data, no labels, restricted networks — and design every POC with a clear path to production.•   Demo & Storytelling: Deliver live demos in Thai and English tailored to the audience — technical deep-dives for engineering teams, business-outcome narratives for executives. Translate model metrics into business impact: OEE uplift, downtime reduction, yield improvement, cost savings.•   Production Delivery & Handover: Harden POCs into production systems — authentication, secrets management, monitoring, CI/CD, containerized deployment — then complete handover with documentation, runbooks, training, SLAs, and an ROI measurement report that proves the value delivered.2. AI/ML Engineering Across Domains (40%)•   LLM & Agentic AI: Build RAG pipelines (chunking strategies, embeddings, vector databases such as pgvector/Qdrant/Milvus, hybrid search, reranking) and agentic systems (LangChain/LangGraph, tool use, structured output, guardrails). Optimize serving with vLLM/TGI and quantization; evaluate with Ragas/DeepEval and LLM-as-judge; handle Thai–English code-switching and Thai tokenization challenges.•   NLP & Document AI: OCR (PaddleOCR/Tesseract/cloud OCR), layout understanding (LayoutLM-family, Donut), key-value and table extraction from invoices, POs, and QC reports — including scanned, mixed Thai–English documents. Speech (Whisper-based ASR fine-tuned for Thai) for call-center analytics and voice interfaces.•   Computer Vision & Image Processing: Industrial defect inspection (YOLO/RT-DETR, segmentation, few-shot anomaly detection with PatchCore/PaDiM), video analytics (tracking, PPE compliance, counting), and classical image processing where it wins — camera calibration, lighting design, sub-pixel gauging. Edge deployment via ONNX/TensorRT/OpenVINO on NVIDIA Jetson and industrial PCs.•   Signal Processing & Time-Series AI: FFT/STFT, wavelets, and envelope analysis for vibration-based condition monitoring of rotating equipment; anomaly detection and forecasting on sensor data; remaining useful life (RUL) estimation; streaming analytics with alert thresholds that operators actually trust.3. Industrial Data & Systems Integration (25%)•   IIoT & OT Connectivity: Read data from PLCs (Siemens, Allen-Bradley, Mitsubishi, Omron) safely and non-intrusively via OPC UA, MQTT (incl. Sparkplug B), and Modbus TCP. Respect IT/OT segmentation and the Purdue model; manage certificates and least-privilege access on the plant network.•   Ignition (Inductive Automation): Work with the tag system and Tag Historian, build operator-facing Perspective dashboards, script in Jython (gateway/tag-change events), and expose REST endpoints via WebDev — writing AI predictions back to tags and surfacing insights where the work happens.•   HighByte & Unified Namespace (UNS): Use HighByte Intelligence Hub to model raw OT data into standardized, contextualized payloads at the edge. Design ISA-95-aligned MQTT topic hierarchies (Enterprise/Site/Area/Line/Cell) on brokers like HiveMQ/EMQX — because a well-modeled UNS is the difference between 2 weeks and 6 months of data prep for every AI project on top of it.•   Backend & Data Pipelines: Build FastAPI/gRPC services and pipelines connecting SCADA, MES, ERP, historians (Ignition Historian, AVEVA PI), and time-series databases (TimescaleDB, InfluxDB) to AI platforms on AWS.Background / Experiences•   Engineering Foundation: B.Eng. or B.Sc. in Computer Engineering, Computer Science, or a related field — with a solid grasp of data structures, algorithms, and system design.•   Production AI Experience: Hands-on experience deploying AI/ML systems into real-world environments — not just research or coursework. You have shipped something that real users or operators depend on.•   Python Proficiency: Strong command of Python with clean, modular, production-grade code — Async I/O, type hinting, testing, and solid OOP applied in practice.•   Client-Facing Mindset: Ability to work closely with non-technical stakeholders, gather requirements, and communicate technical concepts clearly in both Thai and English.•   Field Readiness: Comfort working on-site at client facilities — factory floors, PPE, shift-time constraints — and building rapport with the operators whose buy-in determines whether your solution gets used.Knowledge & Skills•   AI Frameworks: Hands-on experience with LangChain/LangGraph, Hugging Face, PyTorch, and the OpenAI/Anthropic SDKs; strong grasp of RAG and agent architectures.•   Backend & APIs: RESTful APIs and microservices with FastAPI (or Flask/gRPC); proficiency in SQL (PostgreSQL) plus familiarity with vector databases (pgvector, Qdrant, Weaviate).•   Cloud & Containers: Working knowledge of AWS (EC2/ECS, Lambda, S3, SageMaker; IoT SiteWise/Greengrass a plus) and Docker; Kubernetes/K3s experience valued.•   Manufacturing Concepts: Understanding of OEE (availability × performance × quality), downtime classification, SPC/quality control, and the ISA-95 functional hierarchy (Level 0–4) — how SCADA, MES, and ERP interact.•   Version Control & Collaboration: Git for daily development, code reviews, and shared repositories; comfortable in CI/CD workflows.Bonus Points (Preferred if you have)•   Deep Manufacturing Experience: Hands-on work with Ignition, HighByte Intelligence Hub, UNS/MQTT architectures, OPC UA, or plant-floor data systems in discrete, batch, or continuous process manufacturing (e.g., food & beverage, starch/sugar, automotive, electronics).•   Edge AI Deployment: Model optimization and deployment on NVIDIA Jetson or industrial PCs (TensorRT, ONNX Runtime, OpenVINO) with industrial cameras (GigE Vision, RTSP).•   Thai NLP / LLM Work: Experience with Thai-language models, word segmentation, or Thai–English code-switching data.•   Workflow & Observability: Airflow/Prefect for pipelines; Grafana, Langfuse/LangSmith, or OpenTelemetry for monitoring and tracing.•   Local LLMs: Curiosity-driven experimentation running models locally with Ollama or vLLM, including quantization (GPTQ, AWQ, GGUF).•   Craft Signals: Open-source contributions, publications, or side projects demonstrating engineering craft.