Om Bharatiya

Lead Engineer | AI Systems & Distributed Infrastructure

📍 Abu Dhabi, UAE • 📧 [email protected]

Building production AI systems that process 200k+ requests/min. Previously scaled distributed systems serving millions. Open source contributor with 4k+ GitHub stars.

---

What I'm building in the AI era

System	Scale	Achievement
Fintech Document AI	200k docs/min	99.5% accuracy, $0.01/doc (90% cost reduction)
Portrait Generation Pipeline	10k+ cards/batch	95% cost reduction vs commercial APIs
MCP Server Platform	O(log n) complexity	33% infrastructure savings, 57% compute optimization
Database Optimization	45s → 200ms queries	225x performance improvement
System Reliability	88% → 99.9% uptime	$2M+ saved in downtime costs

Core Stack

AI/ML: OpenAI GPT-4, Anthropic Claude, Llama 3.1 • Stable Diffusion XL, InstantID • LangChain, PyTorch, Hugging Face • RAG systems, agentic workflows, vector search

Backend: Java, Python, Go, TypeScript • PostgreSQL, Redis, DynamoDB, Elasticsearch • Kafka, AWS SQS, EventBridge

Cloud: AWS (Bedrock, SageMaker, ECS/EKS, CDK) • Kubernetes, Terraform • NewRelic, DataDog, Prometheus

My work

Agentic Compliance Engine
AI agents that understand regulatory nuance and auto-generate audit trails. Multi-modal validation pipeline with 98% audit completeness, reducing review time from 1-3 weeks to 2-4 hours.

DiffusionID Production Pipeline
Identity-preserving portrait generation at scale. 8 seconds/image on RTX 4090, 450 images/hour throughput. Two-stage pipeline with automated PDF card generation.

FAANG Interview Questions
3.7k+ stars. Comprehensive coding interview prep resource.

Screwdriver CI/CD
10+ merged PRs across UI, models, and core platform. Added GitHub/GitLab PR validations for 10k+ engineers.

TranscriptAI
Multilingual transcription using OpenAI Whisper.

My approach to building AI systems

def architect_ai_system(business_requirements):
    """
    Lessons from shipping AI to production:
    1. Start with the data pipeline (garbage in, garbage out)
    2. Vector databases are the new relational databases
    3. Prompt engineering is systematic, not magic
    4. Always have a fallback to traditional algorithms
    5. Monitor for drift, hallucinations, and edge cases
    """
    if requires_compliance():
        return multi_layer_validation() + audit_trails()
    elif requires_scale():
        return async_processing() + caching_layers()
    return simple_and_maintainable()

Currently

Researching multi-agent orchestration and vector search optimization. Mentoring engineers on distributed systems and AI system design. Writing technical content reaching 80k+ engineers and founders. Giving consultations to AI teams for building better solutions.

---

Let's build something together

Always excited to discuss AI system design, scaling challenges, and distributed systems architecture.

PS: My resume is on LinkedIn, but my real work lives in production systems serving millions of users 🚀