नमस्ते, I am चेतन खपेडिया.
Artificial Intelligence and Data Science Engineer
Building intelligent systems at the intersection of robotics, edge AI, and scalable MLOps infrastructure. From bare-metal assembly to cloud-native inference pipelines.
Latest Posts
Recent technical deep-dives and articles.
Generating Reliable Quiz Questions with a Local LLM
Feb 2026
What it actually takes to make a 7B local model produce exactly N questions, in the right mix, in valid JSON, every time. A custom Ollama Modelfile (`quiz-master`) on top of qwen2.5-coder:7b, a length-aware token budget (~400 tokens per question), a schema-locked prompt with type-mix balancing, and a multi-provider fallback chain (Ollama → Gemini → OpenAI) so the platform stays online when the M4 host is offline. Plus the failure modes — silent truncation, type collapse, off-by-one counts — that any production prompt pipeline has to handle.
Real-Time Control with ROS 2 on Constrained Hardware
Nov 2025
Where the latency actually comes from in a ROS 2 control loop — DDS, executor model, scheduling, and IPC — and what it takes to hold sub-1 ms jitter on a Raspberry Pi 5 with a PREEMPT_RT kernel. Measurements, not promises: cyclictest baselines, executor tuning, and a reproducible setup for deterministic robotic control.
Building a Self-Hosted Production ML Platform
Aug 2025
Architecture notes — not a tutorial. How experiment tracking, the model registry, the inference layer, and observability fit together on K3s; where the seams are; and which tradeoffs are worth making when one engineer has to operate the whole stack. GitOps promotion, canary rollout, and the SLOs that make it boring.
Publications & Pitch Decks
Research papers and enterprise architecture presentations.
Multi-Modal Deepfake Detection System
A 7-model heterogeneous ensemble — one PyTorch audio-visual transformer (Pinpoint, with gated cross-attention over ResNet18 visual + MFCC audio) and six TensorFlow CNNs (EfficientNet-B4, ResNet-50 ×2, VGG-16 ×2, InceptionV3) — fused with weighted voting and reviewer-facing per-frame heatmaps and mel spectrograms. Designed for cross-generator robustness, not benchmark over-fitting.
Agro-Tech ML Pipeline — Architecture Pitch Deck
An architectural pitch for a precision-agriculture ML pipeline that scales from a single rover to a fleet of edge devices. Focus is on the contract between edge and cloud — ingestion, on-device inference, drift-aware retraining — not on a single model.