name: Uday Shankar Gattu
role: AI Engineer @ PraxisPro (prev. Founding AI Engineer @ Galexor)
education: MS Software Engineering Systems @ Northeastern University
location: New York, NY
what_i_do: |
I build production-grade AI systems end to end — multimodal GenAI,
agentic orchestration, RAG, and high-throughput inference — and I
stress-test them before they ship. 100k+ adversarial vectors against
Vision-Language Models in staging, because if your AI hasn't been
attacked in staging, it will be attacked in production.
domains:
- Multimodal Agentic Systems (LangGraph, DSPy)
- Vision-Language Model Safety & Alignment
- Adversarial ML & Red-teaming at Scale
- RAG Optimization & Retrieval Systems
- High-throughput Inference (vLLM, SGLang)|
🔴 A.R.G.U.S. Custom evaluation harness stress-testing VLMs against 100k+ visual jailbreak vectors and prompt injection attacks. Recursive critique loops with multi-LLM evaluation.
|
🎬 BrandAI Multi-Agent Orchestration with Critic-Creator loops. 35% improvement in alignment scores vs single-shot baselines. CLIP embeddings + dense vector search for semantic retrieval.
|
|
🎥 Text-to-Video System Novel RAG pipeline for video synthesis — solved "style drift" by anchoring generation on retrieved assets. Semantic search for style-consistent asset retrieval before diffusion.
|
🏢 Production Systems @ Galexor (2025) Architected research-to-production pipeline with 56% inference latency reduction via SSE + token streaming. Zero-trust infrastructure for SOC2 compliance. Hierarchical state management for long-horizon reasoning.
|
┌─────────────────────────────────────────────────────────────────────────────┐
│ RESEARCH & MODELING │
│ ────────────────────────────────────────────────────────────────────────── │
│ Multimodal GenAI • Vision Transformers (ViT) • CLIP • Diffusion Models │
│ LoRA/QLoRA Fine-tuning • Adversarial ML • RAG Optimization │
├─────────────────────────────────────────────────────────────────────────────┤
│ FRAMEWORKS & ORCHESTRATION │
│ ────────────────────────────────────────────────────────────────────────── │
│ LangGraph • DSPy • LangChain • PyTorch • TensorFlow • Transformers │
├─────────────────────────────────────────────────────────────────────────────┤
│ INFRASTRUCTURE & DEPLOYMENT │
│ ────────────────────────────────────────────────────────────────────────── │
│ AWS (SageMaker, Bedrock) • Azure AI Studio • Docker • Kubernetes │
│ Terraform • Ray (Distributed Training) • Pinecone • pgvector │
├─────────────────────────────────────────────────────────────────────────────┤
│ COMPLIANCE & OPS │
│ ────────────────────────────────────────────────────────────────────────── │
│ Model Evaluation Harnesses • SOC2/GDPR Patterns • CI/CD for ML │
└─────────────────────────────────────────────────────────────────────────────┘
2026 ────────────────────────────────────────────────────────────────────────►
│
├─► AI Engineer @ PraxisPro (New York)
│ Agentic Workflow Systems (Temporal) • Human-in-the-loop Orchestration
│ Voice-Agent Scheduling • AI Security Research
│
2025 ├─► Founding AI Engineer @ Galexor
│ Lead Architect, OSIRIS Platform (0-to-1) • VLM Safety Alignment
│ 100k+ adversarial vectors • 750+ parallel worker eval infrastructure
│
2023 ├─► MS Software Engineering @ Northeastern
│ GTA: Generative AI, NLP, Prompt Engineering
│ Courses: Advanced LLMs, Responsible AI, Cloud Computing
│
2021 ├─► ML Engineer @ Tata Consultancy Services
│ Azure Cloud • Predictive Maintenance • Petabyte-scale ETL
│ 15% reduction in manufacturing downtime
│
2021 ├─► BTech Mechanical Engineering @ GRIET
│
◄────┴───────────────────────────────────────────────────────────────────────
Open to collaborating on AI safety research, red-teaming tools, and agentic systems.
If you're building something that needs to not break — let's talk.