Forward-deployed AI engineer · Madrid
I join your team as a forward-deployed AI engineer.
A forward-deployed engineer (FDE) works inside your company instead of handing over a slide deck. I learn how the work gets done today, pick the tasks where AI pays off, and build them into your systems with evals that show they work.
My day job is AI Engineer at Jupiter, building multi-agent systems and the evals around them. Before that I wrote no-std Rust for liquid staking tokens and Solana observability tooling. Work history · 31 notes
- Reply
- Within 2 working days
- Fees
- Fixed, fiat or USDC/USDT
- Where
- Remote, CET hours
What a forward-deployed engineer does
Sit with the work
I join your channels and calls and learn how the job is done today, from the people doing it.
Pick one task worth it
We choose the workflow where an agent saves real time, and agree how to measure it before I build.
Ship it, then hand it over
I build in your repo and tools, with evals and tracing, and leave your team able to run and change it.
What I work on
AI
Multi-agent systems
Orchestration, tool use, handoffs, and what happens when an agent fails with nobody watching.
Note: Vercel AI SDKEvals and observability
Regression suites so a prompt or model change is measured. Tracing, cost and latency budgets.
Note: Prompt Engineering Best PracticesLLM infrastructure
RAG pipelines, provider routing, streaming UIs with the Vercel AI SDK and Next.js.
Note: Retrieval Augmented Generation
Solana
Programs in Rust
Writing or reviewing on-chain programs, including no-std. Staking and LSTs.
Note: Stake Accounts in SolanaInfrastructure
Validators, RPC, ingestion into ClickHouse and Tinybird, and alerts that fire before users notice.
Note: Solana ValidatorsAgents that sign
Agents that act on chain: simulate first, enforce policy, log every decision, sign last.
Note: Anatomy of Transaction Simulation
How we can work together
Review
2 weeks, fixed fee
I read the code, run the system and talk to whoever owns it. For AI that means agent design, eval coverage and cost. For Solana, program correctness and operational risk. You get findings ranked by risk, a plan, and a 90-minute walkthrough.
Build
4 to 8 weeks, fixed fee
One defined outcome, built in your repo with your conventions. You get the code, the tests or evals that show it works, a handover document and a written update every week.
Advisory
monthly, agreed scope, cancel anytime
An async channel and one call a week. Architecture and PR reviews, build-or-not questions, hiring help. Answers are written down in your tools.
Bars show length on a shared 8-week scale; dashed means ongoing.
A program review is an engineering review, not a security audit. If you need a report you can publish, hire an audit firm. I can help you prepare for it. Fees are fixed, paid in fiat or USDC/USDT. I don't take equity or tokens.
How an engagement runs
Step 1: We have a 30-minute call and you describe the problem.
Step 2: Within a week I send a written proposal with scope, a fixed price and dates. If I'm not the right person, I say so.
Step 3: I do the work in your repo and write up decisions as I make them.
Step 4: I hand over a document your team can run with, and we close with a call.
Who this suits
Good fit
- Teams putting LLM agents in front of users who need them measurable.
- Solana protocol or product teams that need a Rust/TypeScript engineer for a defined piece of work.
- Infra teams instrumenting chains or LLM systems with ClickHouse or Tinybird.
- Service businesses of 10–50 people with one repetitive task to hand to AI. The free assessment is the place to start.
Not a fit
- Token launches, tokenomics, marketing or community growth.
- AI strategy without a practical implementation goal.
- A formal third-party security audit.
- Hourly staff augmentation with no scope.
Questions
Do you work with teams outside Spain?
Yes. Most of it is async in your tools. Calls are scheduled around CET.
Can you work on EVM?
I have shipped to production on Solana, not EVM. I have notes on Solidity, Base and zk proofs and can advise, but I won't take an EVM build.
Which LLM providers and stacks?
Any provider. My recent work is TypeScript with the Vercel AI SDK. Evals and infra follow your stack.
How soon can you start?
It depends on what I'm already committed to. Ask and I'll give you a date.
Will you sign an NDA?
Yes, a reasonable mutual one, before any code is shared.
Tell me about the problem
A few lines is enough. I reply within two working days.
Or email me@pavs.dev or find me on LinkedIn (opens in new tab).