Mushood
Hanif
Not what a model outputs - how the system decides, executes, and holds under load.
about
Inference is easy. Everything else isn't.
Honest where it matters. Available when it's hard.
As a Senior AI Engineer and Founder of Haga, I specialize in building enterprise agentic systems, LLM fine-tuning, and high-throughput inference architecture. My engineering focus centers on what surrounds generative models: multi-agent state orchestration, low-latency streaming pipelines, and GPU memory optimization. Rather than relying on simple API wrappers, I architect production AI infrastructure — from on-prem QLoRA model tuning to sub-200ms real-time decision systems.
Across 7 years at Afiniti and Confiz, I architected a 6-agent LangGraph document processing pipeline cutting manual turnaround by 92% (3 days to <4 hours), deployed streaming fraud detection handling 1.2M+ daily transactions at <180ms p95 latency, and scaled bilingual RAG engines 8× using async FastAPI and FAISS vector search. Today at Haga, I am advancing physical AI by building the independent trust and physics-consistency verification layer for robot learning policies and generative world models.
Inference as a System
Most teams ship inference as a function call. The real questions - p95 latency, 10x load, what happens when a backend goes down - are architecture questions. I answer them before the first model goes live.Physics-Informed Scientific ML
Data-driven physics models aren't data problems - they're structure problems. Ignoring governing equations forces the model to rediscover physics from data it may never have enough of. Embedding PDEs into the objective is what makes sparse data sufficient.What I don't do
- I don't ship AI wrappers dressed as products. Core API calls with a nice UI aren't systems.
- I don't build for its own sake. The system has to earn what it costs to run.
- I don't take off-the-shelf work. If the implementation is a Google search away, I'm not the right person.
experience & focus
Trajectory & Impact.
Building the independent trust, benchmarking, and physics-consistency verification layer for robot learning policies and generative world models.
- Architected haga-core: Mass & friction degradation curves on MuJoCo / Robosuite environments.
- Engineered physical violation detection (teleportation, hovering, impulse) across CogVideoX & CoTracker datasets.
- Shipped public benchmark evidence browser (haga-web) and automated data room for investor diligence.
impact & metrics
Proof, not promises.
featured work
Systems in Production.
View All ProjectsAgentic Document Compliance Pipeline
For global logistics firms, the manual processing and compliance verification of cross-border documentation (Bills of Lading, Commercial Invoices, Packing Lists) is a multi-day bottleneck prone to costly human error. I architected and deployed the Agentic Document Compliance Pipeline, a state-of-the-art, 6-agent orchestration system powered by LangGraph. By fine-tuning LLaMA-3 with QLoRA and routing verifications through a custom Model Context Protocol (MCP) layer, the pipeline cut manual document-processing time by ~92%, raising extraction accuracy to 96% F1 while drastically reducing hallucination and false-approval rates.
Azure GPU-Backed Streaming Inference Migration
Deploying heavy Deep Learning models in production often results in unacceptable latency and bloated cloud compute costs if architectural patterns are not modernized. The Azure GPU-Backed Streaming Inference Migration represents a masterclass in MLOps engineering: the end-to-end migration of a legacy, on-premise CPU batch-inference system to a high-throughput, GPU-backed streaming microservice on Azure Kubernetes Service (AKS). By implementing Server-Sent Events (SSE) via FastAPI, dynamic tensor batching, and automated Blue-Green failover protocols, this migration cut inference latency by 92% and compute costs by 64%.
writing & thoughts
Technical Insights.
Read All ArticlesThe Great AI Content Backlash: Why Audiences Are Rejecting Polished AI Posts in 2026
AI-generated content just overtook human writing online, yet consumers are pulling away from overly polished AI posts. Here's the authenticity backlash reshaping social media in 2026 — and why LinkedIn's AI bet backfired.
Agentic AI in 2026: Why Autonomous Agents Are the Defining Trend of the Year
AI agents that plan, use tools, and act without constant human input have become the dominant story in AI. Here's what's driving agentic AI in 2026 and why reliability is now the industry's top priority.
stack
What I run in production.
Profiled under load. Not just imported.
PyTorch
Deep LearningPrimary framework for neural networks, fine-tuning & custom model pipelines.
Hugging Face
LLM & NLPTransformers, PEFT adapters, tokenizers & open-source model ecosystem.
LangGraph & LangChain
Agentic SystemsMulti-agent state machines, graph-based routing & tool orchestration.
scikit-learn
Machine LearningGradient boosting, classification, evaluation & active learning loops.
Python
Core LanguagePrimary language for production AI infrastructure, microservices & MLOps.
TypeScript
Core LanguageType-safe application engineering, web interfaces & full-stack tooling.
SQL
Data QueryingComplex relational queries, analytical window functions & indexing.
Bash
ScriptingLinux shell automation, deployment scripts & cluster administration.
Git
Version ControlDistributed version control, release branching & code review pipelines.
Linux
OS & KernelEnterprise Linux administration, process management & GPU driver setups.
FastAPI
Async FrameworkHigh-performance REST & streaming endpoints handling 95+ req/s at <400ms p95.
Celery
Task QueueDistributed asynchronous worker queues & background task execution.
Redis
In-Memory StoreCaching, pub/sub messaging, rate limiting & session state.
PostgreSQL
Relational DBEnterprise relational database persistence & ACID-compliant transactions.
gRPC
Inter-Service RPCLow-latency protocol buffer RPC for high-concurrency microservices.
Bun & Next.js
Web & JS RuntimeUltra-fast JavaScript/TypeScript runtime & modern React server components.
FAISS & Vector Search
Vector SearchHNSW dense vector indexing cutting retrieval latency from 340ms to 60ms.
Apache Airflow
Workflow OrchestrationAutomated ETL pipeline DAGs, data movement & scheduled data processing.
pandas
Data ProcessingData manipulation, preprocessing, active learning & evaluation analysis.
Docker
ContainerizationContainerized service packaging, multi-stage builds & environment isolation.
Kubernetes
OrchestrationProduction container orchestration, AKS & GKE cluster scaling.
'%20transform='translate(.587%204.468)%20scale(.91904)'/%3e%3cpath%20d='M97.28%2081.607H37.987a2.743%202.743%200%2000-1.874%204.751l38.1%2035.562a5.991%205.991%200%20004.087%201.61h33.574z'%20fill='%230078d4'/%3e%3cpath%20d='M46.09.002A6.434%206.434%200%200039.93%204.5L.644%20120.897a6.469%206.469%200%20006.106%208.653h32.48a6.942%206.942%200%20005.328-4.531l7.834-23.089%2027.985%2026.101a6.618%206.618%200%20004.165%201.519h36.396l-15.963-45.616-46.533.011L86.922.002z'%20fill='url(%23azure-original-b)'%20transform='translate(.587%204.468)%20scale(.91904)'/%3e%3cpath%20d='M98.055%204.408A6.476%206.476%200%200091.917.002H46.575a6.478%206.478%200%20016.137%204.406l39.35%20116.594a6.476%206.476%200%2001-6.137%208.55h45.344a6.48%206.48%200%20006.136-8.55z'%20fill='url(%23azure-original-c)'%20transform='translate(.587%204.468)%20scale(.91904)'/%3e%3c/svg%3e)
Microsoft Azure
Cloud PlatformAzure ML Studio, GPU streaming microservices & production AKS clusters.
Amazon Web Services
Cloud PlatformEC2 GPU instances, SageMaker model training & cloud infrastructure.
Google Cloud (GCP)
Cloud PlatformGKE Kubernetes clusters, Vertex AI & Cloud Run serverless services.
get in touch
Hard problems welcome.
Heads-down building right now - not looking for roles. But if you've got a hard problem, a wild idea, or just want to talk shop about LLMs, distributed systems, scientific ML, or why this site is unreasonably over-engineered for a portfolio, I'm always up for that.