Work with me
Four things I'm genuinely good at.
Scoped engagements rather than open-ended hours. Each one links to work that proves it — if I haven't shipped it, it isn't on this page.
Custom AI Agent Development
Tool-using agents that do real work in your systems — with confirmation gates on anything irreversible, hard termination limits, and a trace for every run.
- —Agent with your tools wired in and scoped per role
- —Confirmation and audit rules on write actions
- —Tracing and a replayable run history
- —Evaluation set so changes can be measured
RAG & Agentic RAG Systems
Question answering over your documents that stays accurate — reranking, self-correction, and an eval harness that scores retrieval separately from generation.
- —Ingestion and chunking tuned to your corpus
- —Hybrid retrieval with reranking
- —Grounded answers with citations
- —Retrieval eval report on your own questions
Voice & Multimodal Agents
Voice agents that hold a real conversation: transcription tuned to your audio, intent routing with a human-handoff threshold, and latency low enough to feel like a phone call.
- —Call analysis and intent taxonomy before any build
- —ASR tuned and measured on your own recordings
- —Speech-to-speech loop inside a latency budget
- —Escalation path that hands a human full context
AI Platform & Backend Engineering
The service the model lives inside: an API with auth, queues, webhook ingestion that survives retries, audit trails and metrics — built so the AI feature is the easy part to change.
- —Async API with authentication, roles and rate limiting
- —Background processing and webhook ingestion with idempotency
- —Data model, migrations and per-account data isolation
- —Metrics, tracing and structured audit logging from day one
LLM Cost & Latency Audit
A fixed-scope audit of what you are spending and why, with a prioritized plan — local model routing, semantic caching, and right-sizing the model to the task.
- —Per-step cost and latency breakdown
- —Local vs API routing plan with projected savings
- —Semantic caching design
- —Written report with an effort-ranked backlog
How it runs
A call about the problem
Not about the tech. What breaks today, what it costs you, and what success would actually look like.
A written scope
Deliverables, timeline and price agreed before any code. If the honest answer is that you don't need AI here, I'll say so.
Shipped in slices
Something runnable early, then iterations. You see progress weekly rather than at the end.
MAJID
“Innovation distinguishes between a leader and a follower.”
Need an AI system that holds up in production? Tell me what breaks today and I'll tell you what I'd build.