High-Performance Systems · Executive AI Engineering · Production Ready
AI/ML Engineer
AI/ML Systems · Deep Learning Models · Computer Vision · RL Autonomous Agents · Generative AI
Building RAG pipelines, autonomous multi-agent systems, and deep-learning architectures from scratch — no black boxes. 20+ production-grade projects shipped across LLMs, computer vision, and reinforcement learning.
The future of AI isn't about replacing humans — it's about amplifying human intelligence. Machine learning teaches patterns, deep learning reveals hidden insights, and reinforcement learning shows us how to adapt.
Together, they're the building blocks of tomorrow's intelligent world.
— Rushikesh Mohalkar
I'm an AI/ML Engineer specializing in Large Language Models, Agentic Architectures, Deep Learning, and Production ML Pipelines. I build scalable RAG architectures, fine-tune open-weights models (Llama 3, Qwen, Mistral), and construct autonomous AI agents that plan, critique, and execute tool calls with high reliability.
From low-level model optimization to real-time evaluation harnesses and production deployment, I focus on turning complex machine learning research into resilient, real-world software applications.
Bengaluru, India · Remote & Global Ready
Once spent 48 hours straight debugging a loss spike — only to find a index alignment error. AI engineering teaches math, code, and relentless perseverance.
Want the full picture? Grab the résumé or dig through the code.
A selection of AI/ML systems I've designed, trained, and shipped.
A multi-agent AI system for banking accounts-payable that autonomously reads invoices, scores them for fraud and risk, routes approvals, and auto-pays safe transactions — wrapped in six governance layers for full auditability and provenance.
6 governance layers · auto-pay
An enterprise-grade Retrieval-Augmented Generation system for Kubernetes IT operations — hybrid search, cross-encoder reranking, HyDE, corrective and self-RAG loops, and Text2SQL, all orchestrated with LangGraph and guarded by a nine-layer safety framework.
9-layer safety framework
A minimal, highly performant chat model architecture and inference harness built from scratch in PyTorch — designed for low-latency local execution and custom instruction tuning without black boxes.
Low-latency local inference
A RAG-based document intelligence system that ingests PDFs, indexes vector embeddings using FAISS, and performs context-aware question answering with FLAN-T5 and Gradio UI.
FAISS vector retrieval
An AI-powered property tax copilot built with Google Gemini API, Flask, and Tailwind CSS for real-time tax calculations, regulatory compliance guidance, and automated payment deadline tracking.
Real-time tax calculations
An end-to-end LLM fine-tuning workbench and Flask deployment service using Hugging Face Transformers for domain-specific model adaptation, LoRA parameter tuning, and evaluation.
Domain adaptation pipeline
A privacy-first AI chat app that runs Qwen LLMs locally via Ollama — a minimal Python backend, a simple web UI, and optional Docker, for fast offline conversations with no cloud APIs.
100% offline & private
A LangChain-orchestrated agent that analyzes job descriptions, retrieves relevant experience from a resume via local FAISS + sentence-transformers RAG, researches companies with DuckDuckGo search, and generates tailored cover letters and fit scores.
0–100 fit scoring + gap analysis
A complete Retrieval-Augmented Generation system with no LangChain and no vector-DB server — structure-aware chunking, hybrid search, reranking, contradiction handling, and an eval harness.
A Seq2Seq neural machine-translation engine that translates between English and Marathi using full-sentence context rather than word-by-word substitution.
A Flask-based dual NLP engine using DistilBERT for contextual Question-Answering and real-time sentiment classification (positive/neutral/negative) over custom document text.
DistilBERT QA & Sentiment
An automated NLP text summarization workbench supporting BART, T5, and Pegasus transformer architectures for instant document summarization and batch CSV analysis.
BART / T5 / Pegasus suite
An LSTM network in TensorFlow/Keras that reads free-text patient symptoms and predicts a likely disease and a recommended prescription via a dual-output, multi-class softmax head.
A transfer-learning pipeline that classifies chest X-rays as normal or pneumonia using a frozen ImageNet-pretrained VGG16 convolutional base with a dense softmax head.
A Pygame self-driving car that learns to navigate tracks on its own using Proximal Policy Optimization (PPO) deep reinforcement learning.
A real-time computer-vision pipeline that monitors intersections, detects traffic signals and zebra crossings, and flags vehicles that violate stop protocols.
A deep convolutional neural network (CNN) in TensorFlow/Keras with early stopping, data augmentation, and custom convolution layers for binary image classification.
A webcam AR arcade game where your hand becomes the gun — MediaPipe Hand Landmarker tracks index-finger direction to draw an aiming laser, an upward finger-flick fires, and a fist reloads.
A stacked LSTM recurrent network in Keras/TensorFlow that forecasts Google's opening stock price from 60-day input sequences with Dropout regularization.
A PyTorch movie recommender that trains a Restricted Boltzmann Machine on the MovieLens dataset for collaborative filtering using CD-10 contrastive divergence.
0.2544 test MAE
A single visual map of the entire machine-learning landscape — from supervised learning to transformers and diffusion — with the core math on every node and click-through diagrams of how the flagship algorithms actually work.
Deep dives on architectures, research, and the ideas shaping modern AI.
A practical hands-on guide to parameter-efficient fine-tuning (PEFT): 4-bit quantization, Low-Rank Adaptation (LoRA), Memory footprint management, and 2x faster training with Unsloth.
Read articleA comprehensive technical architectural guide to building production multi-agent systems: task decomposition, agent delegation, state isolation, handoff contracts, and failure recovery.
Read articleAn end-to-end technical deep dive into moving beyond naive vector RAG: dense+sparse hybrid search, cross-encoder reranking, chunking strategies, and automated evaluation metrics.
Read articleA deep technical analysis of the shift from pre-training compute scaling to inference-time scaling: Monte Carlo Tree Search (MCTS), Process Reward Models (PRMs), and self-correction loops.
Read articleA technical guide to building agent harnesses: the runtime, state, tools, policies, evaluation, and observability that turn an LLM loop into a dependable system.
Read articleA practical deep dive into production model deployment: packaging, serving, scaling, rollout strategies, observability, and the failure modes that matter after training.
Read articleMy professional path, education, and the skills I'm building on.
Building software pipelines and automating workflows to support reliable delivery across releases.
Building AI/ML systems with a focus on deep learning, reinforcement learning, and production-grade intelligent systems.
Foundation in electronics, communication systems, and software engineering.
Have a project, a question, or just want to talk AI? I'd love to hear from you.
The fastest ways to get in touch — no forms, no scripts, just a message away.