Freelance ML engineer and AI strategy consultant based in Europe, working remotely with teams that need honest answers before they scale models or buy more GPU.

Email · LinkedIn · CV · Google Scholar


AI strategy & readiness

For: leadership and product teams exploring ML or Gen AI without a clear internal owner.

Typical work:

  • “Do we need AI for this?” — problem framing, feasibility, build vs buy, and risk (data, compliance, ops).
  • Readiness assessment: data, team skills, existing tooling, and what a sensible first milestone looks like.
  • Integration roadmap: phased plan from pilot to production, with explicit non-AI alternatives when they are the better fit.

Deliverables: written assessment, prioritized options, and a concrete next-step proposal (often 1–2 weeks).


MLOps audit & production ML review

For: teams with models in staging or production that lack reproducibility, monitoring, or release discipline.

Typical work:

  • Pipeline review: training, evaluation, deployment, rollback, and experiment tracking (W&B, MLflow, Sacred, etc.).
  • CI/CD and containerization gaps; observability and alerting for model/version drift.
  • Recommendations ranked by impact vs effort, with optional hands-on implementation support.

Deliverables: audit report, remediation backlog, and optional follow-on implementation (2–4 weeks common).


Edge perception & real-time computer vision

For: robotics, logistics, and embedded teams shipping perception on constrained hardware.

Typical work:

  • Detection, depth, and fusion pipelines on Jetson-class devices (TensorRT, latency budgets, sim-to-real).
  • Synthetic data strategy, mAP/latency tradeoffs, and deployment hardening.
  • Tech lead or IC delivery on perception stacks already in flight.

Deliverables: working pipelines, benchmarks, and documentation for handoff to your engineering team.


Private Gen AI & on-prem RAG

For: organizations that need document Q&A or agents without public-cloud dependency.

Typical work:

  • Architecture for OCR → chunking → embeddings → retrieval (vLLM, Ollama, on-prem deployment).
  • Evaluation of VLMs/RAG quality, privacy boundaries, and operational runbooks.
  • Strategy for Drizz-style product research or docai.tools-style private knowledge bases.

Deliverables: reference architecture, PoC, or production path depending on scope.


How engagements usually run

PhaseWhat happens
Intro call30 minutes — scope, constraints, whether there is a fit
ScopingFixed proposal: deliverables, timeline, rate (day-rate or fixed-scope)
DeliveryRemote, async-friendly; overlap with your timezone where needed
HandoffDocs, code, or runbooks your team can own

I work as an independent contractor (not agency resale). NDA and EU/GDPR-aware setups are routine.


Background

6+ years across ADAS perception, aerial segmentation, edge robotics (Verses AI, Edge Case Research, Fraunhofer AISEC, TUM). See full CV and PDF resume for roles, metrics, and publications.

To start: vyasdevms@gmail.com with a short note on your team, problem, and timeline.