Abubakar Siddiq — AI/ML Engineer

02 / Career

Experience

Oct 2025 — Present

Core engineer on Skan's proprietary Agentic Execution SDK — runtime components that orchestrate compiled business processes into multi-agent execution graphs, powering 50+ autonomous agents live across every enterprise client deployment.
Forward Deployed Engineer on engagements with Wells Fargo, UNUM, AMUR, and Dell — owning engagements customer-direct, building demos and POCs that converted prospects into production, and delivering agent automations that cut manual process time by ~45%.
Sole owner of the proprietary DSL Compiler service bridging process discovery to the agent runtime — designed the IR, validation passes, and code-generation layer end-to-end.
Productionized the platform on Kubernetes (GKE) with Helm and Dockerized FastAPI microservices; own CI/CD on GCP & AWS with a Qdrant vector DB backing tool-calling, RAG, and agent memory.
Built an agentic evaluation & guardrail system auditing Process Agents pre-deployment; lead LLM selection, per-client fine-tuning, and the observability/tracing layer used to debug agents in production.

LangGraphKubernetesGKEQdrantFastAPILLM Evals

Aug 2024 — Oct 2025

Designed and shipped a multi-agent LangGraph/CrewAI system orchestrating specialized agent roles across client-facing workflows, deployed on AWS ECS with auto-scaling.
Built an end-to-end web-crawling & data-ingestion pipeline that finds, filters, enriches, and packages client-specific datasets into production RAG knowledge bases — custom chunking, deduplication, and Pinecone indexing for low-latency retrieval.
Integrated WhatsApp and Twilio channels via the Model Context Protocol (MCP), unifying conversation state across platforms for continuous multi-channel reach.

CrewAIAWS ECSRAGPineconeMCP

Mar 2024 — May 2025

Built a custom Masked Autoencoder vision model and data pipeline on drone RGB orthomosaics for Leaf Area Index estimation — R² = 0.87 with RGB-only inputs, removing the need for costly multispectral hardware.
Designed a ViT pipeline ingesting DSM/DTM multispectral inputs for plant-height estimation — RMSE 0.26 m, outperforming CNN baselines; built the OpenCV/rasterio/PyTorch preprocessing pipeline used lab-wide.

MAEViTPyTorchRasterioDrones

Aug 2021 — Nov 2021

Built YOLO-based video analytics for UAVs — real-time detection, tracking, and classification from aerial streams, boosting autonomous surveillance coverage by 60%.
Integrated SOTA segmentation models into the DataSync Intelligence pipeline — 40% faster anomaly detection and 25% higher precision in GPS-denied landings.

YOLOSegmentationUAVs

🎓