I didn't start as an AI engineer. I earned it, one production system at a time.

Physics + Data first. M.Tech from IIT (ISM) Dhanbad, petroleum engineering with a minor in ML. My first job was building physics-assisted ML models that worked alongside reservoir simulation - production forecasting, EOR screening, history matching using sequential deep learning. Domain knowledge baked in from day one. That work got published in the Journal of Petroleum Science and Engineering.

Then real engineering kicked in. At Siemens (3+ years), I stopped working in notebooks and started shipping systems. Automotive anomaly detection processing 10TB+ of image data per day with under 5-second latency, built on NVIDIA Triton, MQTT, and concurrent inference pipelines. Smart city drone vision with scalable DL service templates deployed across 10+ use case owners. Dielectric fluid leakage detection on time-series. Satellite imagery segmentation for building boundary analysis. Every project meant understanding a new domain, building the ML, writing the backend service, hooking into enterprise infra, and keeping it running. I also mentored junior developers through the notebook-to-production transition - because I had made that journey myself.

GenAI arrived and I was already production-ready. RAG pipelines, LLM-powered APIs, Azure OpenAI integrations, LangGraph agents. Not a newcomer - someone who already knew how systems actually run. Picked up patents along the way: Federated Auto Vision for automotive defect detection, data encryption systems.

Now at Zensar as ML Lead, running a cross-functional team delivering an enterprise CV platform for weld quality control on H100 GPUs. And on the side - I'm building InterviewValley.com entirely solo. Full voice AI pipeline (STT → LLM → TTS), FastAPI backend, Next.js frontend, MongoDB, Docker, Capacitor mobile app. AI coding tools closed whatever frontend gap remained. From idea to shipped product, alone.

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
  ZENSAR  |  Weld Quality Control AI Platform  (Current - ML Lead)
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
  EfficientNet-B0 + Vision Transformer │ H100 GPU with MIG partitioning
  Real-time feedback loop → retraining pipeline
  React 19 + TypeScript frontend │ FastAPI backend │ Docker Compose
  Led cross-functional team end to end

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
  SIEMENS  |  Automotive Anomaly Detection
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
  10TB+ image data / day │ <5 second processing latency
  NVIDIA Triton Inference Server │ MQTT broker │ Postgres service
  Concurrent DL inference │ Model registry │ CI pipelines
  YOLO-based object detection │ Multi-use-case production deployment

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
  SIEMENS  |  Smart City Drone Vision AI
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
  Scalable DL service template → 10+ use case owners
  Tiny object detection with SAHI │ FasterRCNN │ ResNet
  RabbitMQ orchestration │ Docker-optimised service containers
  Environment compliance for construction in smart city

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
  SIEMENS  |  LLM Minutes of Meeting Tool
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
  Real-time transcription → structured MoM → auto-email to attendees
  FastAPI + WebSocket + MongoDB + Azure OpenAI (private GPT endpoint)

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
  DICELYTICS  |  Physics-Assisted ML for Oil & Gas
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
  Reservoir simulation + ML hybrid models
  Production forecasting │ EOR screening │ History matching
  Published → Journal of Petroleum Science and Engineering

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
  INTERVIEWVALLEY.COM  |  Solo · Full Stack · Shipped
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
  Voice pipeline: Deepgram STT → LLM → ElevenLabs TTS  (real-time)
  FastAPI backend │ Next.js + Tailwind frontend │ MongoDB
  Capacitor → Android/iOS native app │ Docker deployment
  Built entirely solo with AI-assisted development

🤖 ML / Deep Learning / Computer Vision

🧠 GenAI / Agents / RAG

🎙️ Voice AI

🖥️ Backend, Frontend & Data

📡 Infra, Queues & DevOps

Published Papers

• 📄 Data-driven hydrocarbon production forecasting using ML - Journal of Petroleum Science and Engineering
• 📄 GV Black Hierarchical Multiclass Classification · Panoramic Radiographic Synthetic Data - IEEE
• 📄 Beyond Traditional Vibration Classification - ADA Science
• 📄 Comparative study of GANs for sensor data generation & remaining useful life classification

Patents

• 🔐 System and Method for Data Encryption and Decryption
• 🔐 Federated Auto Vision for Automotive Defect Detection and Information Retrieval (Siemens)

Built the habit early. Still doing it.

• YouTube - Petroleum from Scratch: petroleum data analytics, Python, ML from first principles
• Great Learning - sessions on anomaly detection, decision trees, NLP, time series, model evaluation
• Guest Lectures - universities and webinars across India and globally
• Analytics Vidhya / AI Planet - community sessions on practical GenAI and ML
• At Siemens - mentored junior devs from Jupyter notebooks to production ML services

 

📅 Commit Activity

I post what I'm building, what broke, and what I actually learned. No fluff - real process, real code.

Follow the build → @JaiyeshChahar