Arjun Arjun@ArjunMurthy

About me

Track record (10+ yrs): Hands-on architect/tech lead shipping vision + GenAI systems on edge and cloud; ~5 yrs customer-facing for discovery, pilots, and rollouts. Flagship builds: Swift Vision (Vision+GenAI, edge-first): Open-vocabulary defect detection (Grounding-DINO + SAM), change-maps (SSIM/LPIPS); optimized on Jetson Orin / Qualcomm Cloud AI 100 via TensorRT/TFLite → <500 ms median latency, privacy-by-design (OIDC, signed bundles). GenAI Retail Assistant: RAG over governed Delta Lake/Unity Catalog; LangGraph/LangChain tool-graph; Structured Outputs (JSON Schema) + validator gates; on-prem vLLM/TGI → reduced L1 tickets/MTTR with traceable sources. SwiftDecoder (Smart OCR & Locator SDK): ML components with Paddle Lite/Google ML Kit; SDKs for Java/Kotlin, Swift, React Native → 1,000+ devices, 20+ customers; golden-image regression + active learning. Modeling & optimization: PyTorch/TensorFlow, PEFT/LoRA, quantization (FP16/INT8/4-bit), ONNX→TensorRT/TFLite, super-resolution/deblur/OCR rectification. Agentic AI: Routers/planners, tool use, validator-gated escalation (small→large model), prompt-injection defenses, cost controls (caching, cascades). Data & KG: GROBID/Unstructured → NER (scispaCy) → relations (REBEL) → Neo4j/RDF with provenance; boosts retrieval for grounded answers. MLOps & reliability: CI/CD with golden-image suites, eval gates (EM/F1, citation-entailment, p50/p95), canaries/rollback, telemetry, active-learning loops. Security & compliance: OIDC SSO, signed model bundles, tenant isolation, PII masking. Collaboration: Led cross-functional squads (3–10 across Android, CV/ML, cloud, QA); clear stakeholder comms, decision logs, delivery to SLOs. Bottom line: I build reliable, low-latency, on-prem/edge-ready AI—grounded, measurable, and production-hard.

Socials