Skip to content

Services

Senior engineering,from model to production.

Four practices that cover the whole path a production AI system travels: retrieval and agents, vision models, the AWS platform beneath them, and the devices and realtime interfaces at the edge.
01

Multi-agent LLM systems & RAG

Agent systems that do real work and answer with sources. We build the retrieval, orchestration, evaluation, and serving layers, on Amazon Bedrock or self-hosted on your own GPUs.

  • Citation-grounded RAG where every answer cites its page, video frame, or transcript timestamp: BGE-M3 embeddings, Qdrant retrieval, cross-encoder re-ranking, tokens streamed over SSE.

  • Multi-agent orchestration with LangGraph and Amazon Bedrock Agents: crash-resumable pipelines and specialized agents such as chatbot, medical scribe, and risk analytics.

  • Self-hosted inference with vLLM, speculative decoding, and KV/prefix caching, including fully air-gapped deployments when data cannot leave the building.

  • Prompt-injection defence (retrieved evidence treated as untrusted), Bedrock Guardrails, LLM-as-judge evaluation, and visible AI-generated labels.

Amazon BedrockLangGraphvLLMQdrantBGE-M3Bedrock Guardrails
02

Vision & applied ML in production

Vision models taken from dataset to a scale-to-zero endpoint. We handle curation, training, evaluation, deployment, and the app that puts results in front of a user.

  • ViT transfer learning with DINOv3: one model for 10-class classification and pixel-level segmentation in a single forward pass.

  • Dataset curation that fixes the real problem: an 11-source clinical dataset assembled to remove severe class imbalance.

  • SageMaker training through to scale-to-zero async inference endpoints, with MLOps.

  • Measurement and screening around the model: AprilTag calibration for cm²-accurate measurements, Amazon Rekognition pre-screening, Comprehend Medical for ICD-10-CM, RxNorm, and PHI.

PyTorchHugging FaceDINOv3SageMakerRekognitionComprehend Medical

In production

03

AWS cloud & platform engineering

Infrastructure as code, event-driven, and sized for your stage. We design the VPC, compute, identity, and observability so the AI layer has something solid to stand on.

  • Everything in CDK or CloudFormation, delivered through GitHub Actions and Docker.

  • Compute fit to the workload: ECS Fargate, Lambda (including Rust), EC2, and Elastic Beanstalk, fronted by API Gateway, ALB/NLB, and CloudFront.

  • Event-driven backbones: EventBridge, Step Functions, SNS/SQS, Kinesis, and Data Firehose, with Glue, Athena, and Redshift on the data side.

  • Identity, secrets, and audit: Cognito, OAuth 2.0/OIDC, RBAC, TOTP MFA, Secrets Manager, CloudWatch, and HIPAA-aligned VPC design.

AWS CDKECS FargateLambdaEventBridgeCognitoCloudWatch
04

Edge, IoT & realtime systems

Firmware, fleet, and the realtime path to the screen. We write device firmware in C on ESP-IDF and backend services in Rust, secure each device with its own X.509 certificate, and stream telemetry to dashboards and phones.

  • ESP-IDF firmware on ESP32-S3/C5/C6: WiFi CSI capture, 60 GHz mmWave radar fusion, OTA A/B updates, and real-time decisions made on-device, never in the cloud.

  • AWS IoT Core over MQTT/TLS with per-device X.509, feeding Rust Lambda services into DynamoDB, SNS, and EventBridge.

  • Realtime interfaces: SigV4-signed MQTT over WebSocket, Socket.IO, SSE, WebRTC voice and video, and Amazon Chime SDK.

  • Native delivery: SwiftUI iOS with APNs push, and React Native (Expo) with custom TurboModule and Fabric bindings.

ESP-IDFRustAWS IoT CoreMQTT/TLSWebRTCSwiftUI

Engagement models

Three ways to work together.

Every engagement runs under a simple MSA and SOW, invoiced in USD or CAD. Rates on request.
Model 01

Fixed-scope build

A defined deliverable with milestones and acceptance criteria, architected and delivered end to end. You get a finished system, not a staffing arrangement.

Best for

A production RAG or agent system, a vision pipeline, an IoT backend, or an AWS platform with a clear finish line.

Model 02

Fractional / retainer lead

Senior AI and cloud leadership embedded with your team on a retainer, with the time commitment agreed when we scope the work. Architecture, code review, the hardest problems, and hands-on build time.

Best for

Startups and mid-market teams that need senior hands-on judgment before a full-time hire makes sense.

Model 03

Contract augmentation (C2C)

A senior engineer joins your team remotely through GCDC on a corp-to-corp contract for a defined term. You get a senior engineer in your sprints, with no W-2, no visa, and no sponsorship.

Best for

US and Canadian teams filling a Contract, Staff, or Principal AI and cloud role.

Have a system to build? Let’s talk.

Tell us what you are building and where it is stuck. You will get a direct reply, a short scoping call, and a written proposal if we are the right fit.

  • MSA + SOW
  • Corp-to-corp (C2C)
  • W-8BEN-E on request
  • USD or CAD · net-30
  • IP assigned on payment