Mechanical Engineering undergrad at IIT Indore, working across agentic AI systems, applied ML, and production infrastructure.

Worked on automating the software development lifecycle end-to-end — from issue creation through deployment — via a multi-stage agentic pipeline with human checkpoints and automatic error recovery built in. Designed a retrieval layer backed by a graph-based knowledge index so agents could ground their output in accurate, current context. Built a testing framework that generates behavior-driven and end-to-end test suites automatically, with parallelized execution and managed test data. Also worked on the surrounding service architecture and observability.

React FastAPI PostgreSQL Kafka Vault Docker Kubernetes NetworkX New Relic

Researched demand forecasting for logistics networks operating with limited historical data, using hierarchical Bayesian models that share statistical strength across locations to handle cold-start regions. Extended this into a forecast-to-dispatch system, feeding probabilistic forecasts into a rolling-horizon dispatch optimizer that combines classical routing heuristics with graph neural network–based reinforcement learning solvers. Separately modeled vehicle routing under electric-vehicle range and charging constraints.

Python Bayesian hierarchical modeling GNNs Reinforcement learning Vehicle routing optimization

Inter IIT Tech A market intelligence platform pairing regime-aware time-series forecasting with portfolio risk optimization, plus a retrieval-augmented research pipeline over filings, earnings calls, and news for signal generation.

Chronos SARIMAX Kafka AWS PostgreSQL MCP orchestration Graph-RAG

A collaborative robot controlled by voice and gesture: LLM-driven command parsing, vision-based perception for pick-and-place, and a wearable glove that classifies actions from live sensor streams and calibrates itself online to new users.

Llama 3 Groq YOLO Depth perception Hoeffding Trees ESP32 MQTT

Retrieval-augmented code assistant for competitive programming: FAISS similarity search over an embedded problem corpus, feeding an ORPO-finetuned DeepSeek-Coder model for solution generation and explanation.

RAG FAISS ORPO finetuning DeepSeek-Coder

AI-driven custom skin generation for Clash of Clans using Stable Diffusion and ControlNets, with a finetuned CompVis model for style adaptation and lllyasviel's control-v11p-inpaint for prompt-to-image generation.

Stable Diffusion ControlNet Diffusers

Gesture classification from EMG signal data (NinaPro DB5, MYO armbands), with a full preprocessing pipeline (rectification, filtering, smoothing) feeding an attention-based classification model.

Signal processing Attention models PyTorch

• Dhada, Huang, Pendse, Puchkova, McFarlane — Operational Value of Probabilistic Demand Forecasts: A Rolling-Dispatch Case Study (SSRN working paper)
• Dhada, Giaretti, Macias-Aguayo, Bull, Pendse, Girolami, McFarlane — Bayesian Hierarchical Autoregression Model for Cold-start Forecasting (SSRN working paper)

Languages

Infra

ML

• LinkedIn: Varad S. Pendse
• Email: varadpendse.vp@gmail.com