I scope with the customer, build agents, RAG and automation against their real data and tools, then harden and hand over. This page is the engineering profile: three case studies with their evals and limits, the public code, and my CV.
Dubai, UAECustomer-facing engineeringTypeScript · PythonAgents · RAG · Automation
What is a Forward Deployed Engineer?
A Forward Deployed Engineer works inside the customer's environment: finds the real workflow, ships a working system against live data, hardens it and hands it over. Engineer first, customer-facing second. What the job involves · FDE vs Solutions Engineer
Three builds, with their limits.
Each one is labelled for what it is. None of them is presented as a client deployment, and each says what it doesn’t prove yet.
- Open source · public repo + live demo
CiteRAG
Python · FastAPI · LlamaIndex · evalsCitation-grounded RAG with the evaluation built in: hybrid retrieval, per-citation verification, and a golden set that runs against the hosted API.
- What I owned
- retrieval pipeline, citation verification, eval harness, API, deploy.
- Trade-off worth asking about
- Each citation gets its own LLM verification pass. That adds latency (it's on the known-limits list) in exchange for catching citations that don't support the answer.
- Known limits
- One in-memory index per process; a restart loses it unless auto-seed or an external store is used. Citation verification runs as sequential LLM calls, so it adds latency. The golden set covers a small demo corpus, not a client's documents.
- Production boundary
- Needs a persistent vector store, parallel verification and a golden set built from real client questions.
- In development · live demo
SupportPilot
Next.js · Supabase · RAG · approvalsAn AI support workspace: cited answers from help-centre content, a human approval queue for risky drafts, golden-question evals and an audit log on every AI decision.
- What I owned
- product, RAG pipeline, approval workflow, eval runner, widget.
- Trade-off worth asking about
- Production mode fails closed. If auth, rate limiting or bot verification isn't configured, it refuses to serve rather than fall back to demo behaviour.
- Known limits
- No live customer traffic yet, so there are no deflection or satisfaction numbers to report. Multi-user RLS is tested locally; live multi-tenant proof still needs a clean project.
- Production boundary
- A controlled pilot with one support team, measured over a fixed window, before any outcome claims.
- Hackathon build · AMD Developer Hackathon
Autopsy
Multi-agent orchestration · AMD MI300XSix specialised agents per mode across four modes research in parallel, challenge each other, then a synthesiser writes one verdict in about 22 seconds.
- What I owned
- agent roles, orchestration, debate and synthesis flow, UI.
- Trade-off worth asking about
- The whole debate runs on one MI300X in about 22 seconds. Verdict quality has no benchmark yet, and the case study says so.
- Known limits
- No benchmark of verdict quality against expert analysis; it shows orchestration, not research accuracy.
- Production boundary
- Would need a scored evaluation set of known company failures before its verdicts could be relied on.
Public code you can read.
- rag-citation-tool
CiteRAG: golden set, live eval script, known limits in the README.
Open source · public repo + live demo - supportpilot-demo
Support workspace with approvals, RLS policies, rate limits and scheduled evals.
In development · live demo - agent-fleet-audit
Agent production-readiness demo that separates demo state from the production gates it would need.
Working demo · concept + deterministic demo - webflow-agent-kit
TypeScript, Zod-validated agent tools for Webflow. MIT-licensed, public beta.
- autopsy
Multi-agent research and debate across four modes.
Hackathon build · AMD Developer Hackathon - conductor-app
One change request applied across a repo fleet, each repo in its own framework idiom.
Hackathon build · IBM Bob Hackathon - AgentPayOps
Spend-policy control pattern for autonomous agents: approve, escalate or block, with an audit log.
Hackathon build · Vultr + Gemini - Mindoor
Policy layer in front of the model for a clinic front desk, with incident export.
Hackathon build · TechEx Hackathon
- Forward deployed delivery
- Customer-facing scoping, Technical discovery, Solution design, Integration engineering, Customer handoff and enablement, Documentation, Stakeholder communication
- AI & automation
- AI agents, Agentic workflows, RAG, Prompt and context engineering, LLM evaluation, Workflow automation
- AI tooling
- Claude, OpenAI, Gemini, DeepSeek, MCP, n8n, Make, Zapier, Pinecone, pgvector
- Engineering
- TypeScript, JavaScript, Python, React, Next.js, APIs, Web application development
- Web platforms
- Webflow, Sanity, WordPress, Shopify
Customer delivery track record
10+ years shipping software for clients, most of it customer-facing: agency and startup delivery, marketplace products, and now applied AI. Earlier I hired and led a 7-person tech team on a teacher marketplace (the company raised $2M during that period) and led delivery of a sports booking platform that reached 10K+ bookings in year one.
Independent AI Engineer & Web Developer
Freelance · Dubai · 2025 – Present
Agents, document RAG and n8n automations for clients, plus the public builds on this page.
Co-Founder & Web Design Lead
Flowmarc Studio · Dubai · Dec 2023 – Apr 2025
Agency and startup delivery; CRO and technical SEO work; white-label builds with partner agencies; AI and Zapier/Make folded into delivery.
Web App Design & Development Lead
911 Teachers · Dubai · Jan 2020 – Oct 2023
Teacher marketplace (React.js, Flutter); hired and led a 7-person tech team; the product org raised $2M during this period.
Full experience, education, and certifications → /resume
Common questions.
What does a Forward Deployed Engineer do day to day?
A Forward Deployed Engineer scopes real customer problems, writes production code in the customer's environment, debugs integrations (data, auth, APIs), ships systems operators can run, and documents handoff so the work survives after the engagement.
What does Anil Pervaiz specialise in as an FDE?
Agents, citation-grounded RAG and automation on modern web stacks, with the unglamorous parts done properly: evals, human approval paths, audit logs and a written production boundary for every build.
Where can I see proof of work?
The three case studies on this page, the public repositories at github.com/anilandcode (CiteRAG's golden set and eval script are a good place to start), and the CV download below.
Take it with you.
Forward Deployed Engineer
Download my CV
Hiring for an FDE or Applied AI role? Get in touch or email hello@anilpervaiz.com.