Integrations

Selected systems, pipelines, and libraries highlighting ML infrastructure and engineering depth. Focus is entirely on model serving latency, quantization performance, vector retrieval, and operational stability.

• ApplyRail

  Job portal matching 50K+ postings to international students using Gemini embeddings + pgvector; serves from a single Proxmox node at ~180ms p95.

  Flask Postgres pgvector Gemini Docker
  180ms

  p95 Latency
• KernServe

  Custom LLM inference runtime optimized for 6GB GPUs utilizing custom CUDA-accelerated kernels, dynamic batching, and INT4 quantization to achieve 45+ tokens/sec on local consumer hardware.

  Python CUDA C++ GGUF PyTorch vLLM
  45+

  tokens/sec
• RAG-Eval

  Enterprise RAG observability platform calculating semantic similarity, retrieval precision, and answer faithfulness metrics in real-time for Mastercard Assistant Platform (MAP), handling 1.2M monthly queries.

  Python Databricks Splunk Datadog SageMaker
  1.2M

  queries/mo
• Bio-Segment

  Edge-deployed insect mortality detection model running on GCP IoT Core and SageMaker pipelines, automating computer vision analysis of 100K+ daily farm images with an 18% harvest yield gain.

  AWS GCP Step Functions Airflow BigQuery Dataflow
  18%

  yield gain