Varun Rao
Absurd AIML Engineer, Co-founder of Human Slop
About
AI ate my Creatine
Started as a Data Scientist wrangling messy datasets and building
predictive models,
But soon I realized that my true passion lies
beyond just data and more towards AI systems and Deep Learning. And
worked on a wide range of projects for hackathons, competitions,
personal research, and Internship. Got fed up Building in AI all the
time hence Co-founded Human Slop (Anti-AI Social Platform)
Work Experience
Contributing to Shipd, the platform behind frontier AI training data. Solving hard software engineering and ML challenges that stump agents and humans alike, producing the data that pushes LLMs forward.
Led 8-member engineering team to architect production-grade multi-agent financial AI system combining real-time streaming, reinforcement learning, and explainable LLM reasoning. Designed complete 7-layer architecture with temporal data fabric, hybrid ML/RL forecasting pipeline, agentic debate framework, and real-time risk engine. Delivered end-to-end platform integrating live market/news/social data with causal knowledge graph, achieving ~20% returns with 5–8% max drawdown over backtested and paper-traded horizon.
Developed multi-agent automation systems and secure REST APIs integrated with AWS and GCP, reducing manual effort by 40% and increasing throughput by 30%. Built cloud infrastructure maintaining 99.9% uptime through CI/CD pipelines. Optimized performance and reduced infrastructure costs by 25% through proactive monitoring and risk assessment.
Education
Coordinator, DSAI Club — organized hackathon and workshops promoting
AI-driven innovation across 200+ students. Led hands-on ML sessions
for applied ML and research-oriented projects.
Relevant Coursework: Data Structures & Algorithms,
Machine Learning, Natural Language Processing, Deep Learning,
Database Management Systems, Statistics, Operating Systems
Featured Work
It WorkedPrivacy-first social platform that blocks 100% of AI-generated content using hardware-bound biometric authentication and real-time typing forensics.
Production-ready PyTorch library for Deep Mutual Learning enabling collaborative neural network training where multiple networks learn from each other's predictions.
Autonomous Financial Intelligence Platform. Multi-agent AI system for market analysis and portfolio management. Backtested returns: ~20%.
Academic Projects
Spatial-Temporal Graph Neural Network for predicting electron oscillation dynamics with high spectral fidelity in RT-TDDFT simulations.
Designed a modular experiment framework combining PPLM steering, a lightweight RLHF proxy, and hybrid inference-time control.
Intelligent agent-based game with state-based behavior (search, chase, evade) and A* pathfinding.
Spatio-temporal analysis and visualization of network signal distributions across building wings to optimize coverage placement.
Graph Neural Network pipeline for anomaly detection in e-commerce graphs using GIME for learning and GAT for classification.
Disk-backed B+ tree storage engine in C++17 with an LRU buffer pool and write-ahead logging for crash recovery.
Personal Blades
GPU-native vector database in C++/CUDA targeting sub-millisecond search and 100K+ QPS for production RAG systems.
Open-source evaluation engine for ML model health. Detects calibration mismatch, adversarial fragility, and blind spots.
Production-grade multi-agent medical RAG system using GPT-4, LangGraph, and CRAG with hybrid BM25+vector retrieval.
Memory-efficient KV Cache implementation reducing usage by 60-80% for production LLM deployment.
Statistical text watermarking using Plug-and-Play Language Models for imperceptible watermark embedding during inference.
GPU-accelerated agent-based macroeconomic simulator in JAX: households, banks, contagion, and shocks.
Open Source · 19 merged upstream PRs
Reviewed and merged by other maintainers.
Robotics data SDK by Hebbian Robotics. sha256 snapshot receipts,
hflow verify, resumable LeRobot imports into
S3/GCS/Azure, and hardening against hostile paths and SQL.
Silent-failure detection for AI agents. pytest plugin now watches every LangGraph entry point, plus a strict CI mode.
Build-only multi-arch (amd64/arm64) Docker Buildx CI job for the OpenBao migrations image.
SEO meta descriptions across 46 kornia doc modules; shap fixes for waterfall label cutoff and LinearExplainer link.
Technical Arsenal
AI/ML/DL
GenAI & RAG
Systems & Engineering
Core Stack
Achievements & Recognition
Some Highlights to Blabber about of me
Ranked 36/1,711 teams (top 2%) building stable commodity return forecasting models.
Ranked 34/386 participants forecasting Ethereum volatility with high-frequency data in 2 weeks of intense competition.
12 in HFlow (robotics data integrity), 4 in ARGUS (agent reliability), 1 in NVIDIA nvcf, 2 in kornia.
Competed among thousands of participants nationwide building a robust VLM-based solution.
Led team to improve baseline ML results by 23% under tight computational constraints.
Publishing my random deep-dives on Medium covering tech fun rides.