Our projects

Engineering excellence, shipped.

Explore our portfolio of engineering excellence. From simulation to reality — FEA, CAD, robotics, and rapid prototyping.

Selected work

Engineering excellence, shipped.

Explore our portfolio across FEA, CAD, robotics, applied AI, and rapid prototyping. Each build bridges digital design and physical reality.

Simulation · FEA & CFD

IP67 Passive Cooling for 5G IoT Gateway

Optimized the thermal architecture for a fanless 15W outdoor gateway. Performed Conjugate Heat Transfer (CHT) analysis to validate natural convection efficiency.

Simulation · FEA & CFD

Solar Structure Engineering Validation

Structural integrity validation for a rooftop solar mounting system under cyclonic wind loads (150 km/h). Conducted Static Structural Analysis.

Simulation · FEA & CFD

Cleanroom Ventilation Strategy

Airflow distribution simulation for an ISO Class 7 Cleanroom. Analyzed velocity streamlines and pressure gradients to identify dead zones.

CAD · Mechanical

Parametric Automotive Braking Systems

Equation-driven master CAD template for high-performance braking systems with patent-defined ratios and tooling-ready outputs.

CAD · Mechanical

IP67 Industrial IoT Gateway Enclosure

Multi-part injection molding assembly with hygienic design, thermal management, and NEMA 4X environmental sealing.

CAD · Mechanical

Modular Server Rack & Thermal Cabinet

Precision sheet metal assembly with K-Factor calculations for SS304 construction and manufacturing-ready documentation.

CAD · Mechanical

3D-Printed Robotic Actuator Housing

Topology-optimized modular design with print-in-place compliant mechanisms for DfAM innovation.

Robotics · ROS 2

6-DOF Bilateral Teleoperation System

Master-Slave teleoperation architecture controlling S101 arm in USA from India with real-time kinematic mapping.

Robotics · ROS 2

80 kg Payload Autonomous Mobile Robot

Heavy-duty AMR with skid-steering drive, Nav2 navigation stack, and LiDAR-based SLAM mapping.

Robotics · ROS 2

High-Torque Integrated Smart Actuator

Custom servo motor with NEMA 23 stepper, AS5600 encoder, and ESP32-based closed-loop PID control.

Applied AI · ML

Vision-Guided 6-DOF Bin-Picking Cell

YOLOv11-Seg + 3D pose estimation on a Jetson Orin Nano. Picks randomly oriented SKUs from a cluttered bin at <80 ms end-to-end latency with 99.2% detection accuracy.

Applied AI · ML

On-Device Defect Detection Camera

Custom PCB with ESP32-S3 + OV5640 camera running a quantized INT8 CNN in TFLite Micro. Zero-cloud anomaly flagging on a production conveyor — 18 FPS at 0.4 W.

Applied AI · ML

ML-Surrogate FEA Solver for Topology Studies

A physics-informed neural network trained on 12,000 ANSYS runs. Predicts von-Mises stress fields in ~40 ms vs. a 22-minute full solve — 500x faster early-stage design exploration.

Applied AI · ML

AI-Perception Warehouse AMR Fleet

Multi-robot 80 kg AMR fleet with semantic SLAM, dynamic obstacle classification, and centralized task allocation. ROS 2 Nav2 + YOLO + Foxglove dashboarding.

Applied AI · ML

Natural-Language CAD Copilot (SolidWorks)

LLM agent wired into SolidWorks API. Engineers say "thicken rib by 2 mm, then re-run static" — the copilot edits the parametric tree, triggers the FEA run, and returns a diff.

Applied AI · ML

Aerial Swarm Perception Stack

Onboard YOLOv11 object geolocation across a 4-drone swarm. Ground station fuses detections in real time for aerial infrastructure inspection and survey.

Applied AI · ML

RAG Copilot over 40k Engineering Documents

Production retrieval-augmented generation system over a manufacturer's 40,000-document corpus — datasheets, GD&T drawings, internal standards. Hybrid search (BM25 + bge-large-en-v1.5 embeddings) with reranker, agentic query decomposition, and source-cited answers. Cuts engineering-question turnaround from 35 minutes to 22 seconds.

Applied AI · ML

Multimodal GenAI for Automated Inspection Reports

Vision-to-text pipeline that ingests inspection photos plus measurement data and drafts customer-ready reports in under 8 seconds. Built on a fine-tuned Qwen2-VL-7B with a domain-specific reasoning prompt, deployed on a single L4 GPU. Replaces 90 minutes of QA writeup per asset.

Applied AI · ML

LoRA Fine-Tuned CLIP for Industrial Part Recognition

Parameter-efficient fine-tuning of OpenCLIP ViT-L/14 with LoRA adapters (r=16) on 18,000 labeled SKU photographs. Achieves 97.4% top-1 against a 78.1% zero-shot baseline. Adapter is 84 MB versus a 1.7 GB full fine-tune. Trained on a single RTX 4090 in 11 hours.

Applied AI · ML

Feature Engineering Pipeline for Predictive Maintenance

End-to-end feature pipeline on 320 vibration sensors streaming at 12 kHz. Time, frequency, and time-frequency features (RMS, kurtosis, FFT bands, wavelet packets) materialized to a Feast feature store. XGBoost classifier hits 0.94 AUC on failure-prediction with a 72-hour warning horizon — saved ~₹2.1 crore in unscheduled downtime in the first year.

Applied AI · ML

MLOps Platform for 600-Device Edge AI Fleet

End-to-end MLOps platform managing 600 Jetson-deployed inspection cameras across 14 sites. MLflow registry, signed-OTA model rollouts, canary deployments, automatic rollback on accuracy drift, and a Foxglove-style fleet dashboard. Median model deploy went from 9 days to 38 minutes.

Applied AI · ML

RAG Support Copilot for a B2B SaaS Knowledge Base

Customer-support RAG agent over a SaaS platform's docs, changelog, and historical Zendesk tickets. Hybrid retrieval, query-rewriting agent, and grounded citations. Deflected 41% of L1 support tickets in the first quarter — equivalent to 3.2 FTE of support engineer time.

Applied AI · ML

Time-Series Forecasting for Plant Energy Management

24-hour ahead load forecast for a 9 MW manufacturing plant. TFT (Temporal Fusion Transformer) ensemble with engineered weather, shift-pattern, and machine-utilization features. MAPE 3.1% versus 8.7% on the legacy ARIMA baseline — moved annual energy spend down by 6.8%.

Let's build

Have a machine to build? Let's scope it together.

Tell us about your project. We'll respond within 1-2 business days with a preliminary scope and timeline — no boilerplate, no up-sell.