Rishi geeky-rish

I build data pipelines, backend systems and efficient AI, and I measure all of it.

Tip

Open to collaborating on open-source, data engineering and AI systems projects.

🔭 Building now

DataForge: data pipeline orchestration
AssetSense: predictive maintenance
AdaptiLLM: adaptive on-device inference

🌱 Exploring

Spark-based ETL
LLM quantization and edge inference
Model deployment and MLOps
Distributed systems

⚡ Live from GitHub

Refreshed every 6 hours by a GitHub Action in this repo. No third-party servers.

Latest activity

⬆️ Pushed to DailyForge (main) · _{18 Oct 2026}
🔀 Merged PR #6 in Labs · _{18 Oct 2026}
🔀 Opened PR #6 in Labs · _{18 Oct 2026}
⬆️ Pushed to Rishi-Kulkarni-ml-zoomcamp-2026 (main) · _{16 Oct 2026}
⬆️ Pushed to profile (main) · _{14 Oct 2026}

🚀 Featured projects

⚙️ DataForge

Data pipeline orchestration platform: dataset ingestion, Spark transformations, validation and async job execution.

FastAPI Apache Spark MinIO PostgreSQL Docker

Reusable ETL workflows
Schema enforcement and validation
Object storage with metadata tracking
Containerized service orchestration

📡 AssetSense

Industrial monitoring that turns real-time sensor telemetry into anomaly alerts and equipment health scores.

ESP32 C++ MQTT Node.js MongoDB React

Vibration, temperature and current sensing
Anomaly detection and health scoring
Remaining Useful Life (RUL) prediction
Live telemetry dashboard

🧠 AdaptiLLM

On-device LLM inference that adapts its execution settings to query complexity and device constraints.

Android Kotlin C++ JNI TFLite Qwen2.5

Adaptive CPU threads and context length
Query classification for workload profiling
Latency, energy and quality-aware policies
Evaluated across 720 inference runs

🔬 Research

Practical machine learning, reinforcement learning and resource-aware AI.

Reinforcement Learning Time-Series LLM Optimization

Dynamic pricing with reinforcement learning
Efficient on-device model execution
Data-driven system performance evaluation

How AdaptiLLM adapts (diagram)

flowchart LR
    Q([User query]) --> C[Query complexity<br/>classifier]
    C --> P{Adaptive policy}
    P -->|CPU threads| R[On-device Qwen2.5<br/>Android · JNI · C++]
    P -->|context length| R
    R --> M[Latency · energy · quality]
    M -.feedback.-> P

🏆 Recognition

🥇 Best Paper Award, ComSIA 2026 (Springer): first author of Reinforcement Learning for Dynamic Pricing
🌍 Global Public Rank 15: WiDS Datathon Challenge 2024