CASE / 01
LLM systems · Retrieval
Knowledge-grounded AI that earns user trust
Hybrid search, reranking, structured citations, and continuous evaluation—designed as one observable production system.
RAGInformation retrievalEvaluation
Available for remote roles & select projects
Senior Machine Learning Engineer
I build ML systems that survive contact with production.
I design and ship reliable LLM systems, RAG pipelines, AI agents, and evaluation infrastructure—turning ambitious AI ideas into software people can actually trust.
rag.pipeline / prodLIVE
Request trace
req_7F2A9p95 latency842ms
01
Understandintent + constraints
done02
Retrievehybrid search
done03
Rerankcross-encoder
running04
Generategrounded output
done05
Evaluatequality + safety
doneEval passed42 / 42 checks
System healthyall services online
Focused on the hard parts of applied AI
LLM systemsRAGAI agentsEvaluationML infrastructure
01 / Selected work
Systems, not demos.
Representative production problems I solve—from retrieval quality and agent reliability to the infrastructure that keeps models useful.
CASE / 02
Agents · Automation
Agents with tools, memory, and guardrails
Stateful workflows that know when to act, when to verify, and when a human should stay in control.
LangGraphTool useObservability
CASE / 03
ML systems · Delivery
Infrastructure that keeps models dependable
Tested APIs, deployment pipelines, monitoring, and feedback loops that turn model behavior into an engineering discipline.
FastAPIMLOpsMonitoring
02 / Expertise
Across the full
intelligence stack.
The model is only one component. I connect data, retrieval, orchestration, evaluation, APIs, and monitoring into a system that performs outside the notebook.
LLM application architecture
Production patterns for context, memory, structured outputs, caching, and cost-aware inference.
RAG & information retrieval
Chunking, hybrid search, embeddings, rerankers, metadata, and grounded answer generation.
Agentic systems
Tool-calling workflows, state machines, human checkpoints, retries, and safe execution boundaries.
LLM evaluation
Golden datasets, retrieval metrics, quality rubrics, regression suites, and online feedback loops.
Model delivery
FastAPI services, containers, async workloads, versioning, monitoring, and deployment automation.
Applied ML & NLP
Problem framing, experimentation, representation learning, classification, and pragmatic model selection.
Tools I reach for
PythonPyTorchFastAPILangGraphLangChainTransformersVector DBsDocker
03 / Ways to work together
Two goals.
One engineering standard.
For hiring teams
A senior engineer for your remote AI team.
I work best with teams building serious AI products—where strong engineering judgment matters as much as model knowledge.
- Own ML features from problem framing to production
- Collaborate async across product and engineering
- Raise the bar on reliability, evaluation, and delivery
For product teams
Focused help shipping an ambitious AI project.
Bring me a brittle prototype, a retrieval problem, or a product idea that needs a pragmatic path to production.
- Production RAG and AI-agent builds
- Architecture, reliability, and evaluation audits
- Technical direction for small, high-velocity teams
AS
Lahore, PKRemote / Global
04 / About
I care about the gap between an impressive demo and a dependable product.
I'm Abu Sufyan, a Machine Learning Engineer with 6+ years across software engineering and applied AI. My work centers on LLMs, retrieval, agents, and the systems discipline required to run them well.
My foundation includes graduate research in machine-learning-based malware detection for IoT and experience teaching Python. That mix shaped how I work: understand deeply, explain clearly, and ship carefully.
01Evidence over hype
02Simple before clever
03Evaluation from day one
04Own the outcome
How I work
Clarity before complexity.
01Frame
Define the user problem, constraints, and success signal.
02Build
Ship the smallest architecture that can prove the value.
03Evaluate
Measure quality, failure modes, latency, cost, and safety.
04Harden
Add observability, feedback loops, and operational resilience.