Applied AI · Manufacturing Systems
Vinay Gandhi
Applied AI Engineer — Industrial Intelligence
15+ years on the manufacturing floor — now building the AI systems that run there. I connect PLC/IIoT streams, real-time data pipelines, and agentic AI into factory intelligence platforms where decisions have physical consequences.
Building unified factory intelligence — from PLC/IIoT ingest through streaming ML and agentic AI to real-time operator decisions.
Factory Intelligence OS
Five layers. One system. All projects feed into it.
Selected work
Two flagship platforms and their components — full catalog on GitHub.
-
AEGIS
● Live buildFactory AI platform — ingests PLC/MES telemetry in real time, detects production anomalies via streaming ML, and surfaces operator guidance through an LLM copilot. The intelligence layer of the Factory Intelligence OS.
-
QualityMind-RAG
Data + AIConversational analytics over manufacturing data — Hybrid RAG (Pinecone) plus Text-to-SQL, orchestrated with LangGraph agents.
-
ClaimLens
NLPWarranty-narrative NLP for field-quality RCA — macro-F1 0.90 on a 1,200-narrative synthetic eval, Pareto trends for 5-Why / 8D.
-
AutoClaim-VLM
VLMDamage intelligence — raw photo to schema-valid JSON in <90s. Glue/PySpark → S3 → Redshift + DynamoDB, QuickSight BI.
-
Factory Genius
Maintenance AIPredictive-maintenance copilot — surfaces downtime drivers and next-action guidance from plant event streams.
-
Automotive visual QA
VisionEdge CLIP + AWS serverless vision pipeline with MES-style FastAPI integration for inline classification.
-
Edge telemetry plane
StreamingMQTT → Rust edge (NATS) → gRPC → Go cloud (Kafka → TimescaleDB) — the real-time ingest layer under AEGIS.
-
TeleOp Flywheel
● Live demoAttributes exact dollar value to human-in-the-loop teleoperation data and surfaces operator quality at per-session granularity — the data-economics layer of the robot data flywheel.
Factory Systems AI
AEGIS ingests PLC + MES data in real time, correlates signals, and surfaces ML-detected anomalies. The edge-telemetry-plane (Rust/Go · NATS · Kafka · TimescaleDB) is its ingest layer.
Quality Systems AI
A connected quality loop: ClaimLens turns warranty narratives into failure trends (F1 0.90), QualityMind-RAG consumes them for 5-Why / 8D RCA, and AutoClaim-VLM routes damage by severity.
About
AI, data, and the systems that connect them.
- AI / ML
- Agentic RAG · LangGraph · LLM agents · Claude API · NLP · computer vision / VLM · evals · PyTorch
- Data & analytics
- Advanced SQL · Python · ETL pipelines · streaming ingestion · ClickHouse · PostgreSQL
- Systems
- MES · IIoT · PLC · Kafka · NATS · Rust/Go edge · streaming infra · EOL test & conveyance
Next step
The deep dive, or the full catalog.