++++
need to find something fast?

Atharva
Mendhulkar.

Interactive Porygon mascot

Systems Engineer & AI Researcher.

Latest Blogs by Atharva Mendhulkar

SaaS

floework

Enterprise-grade collaborative execution platform featuring zero-lock OCC, SQS FIFO workers, DAG dependency cycle detection, and decoupled Multi-AZ AWS infrastructure.

Multi-AZ AWS
Cloud Infra
240 (100%)
Automated Tests
Zero-Lock OCC
Concurrency
< 15ms
Worker Latency
SaaSAWS Multi-AZDistributed SystemsCloud ArchitectureReal-Time
Live · AWS Certified →
AI Systems

AVARA

Always-on runtime control plane for autonomous AI agents — enforcing intent alignment, ephemeral IAM, provenance, and human-in-the-loop safety at execution time.

8 guards
Security layers
< 5ms
Runtime overhead
100%
Audit replayable
3 lines
Integration
AI SecurityAgent GovernanceRuntime SafetyLLM
Open Source · Production-Ready →
Systems

UPRS

Proactive OS-level daemon operating in the 0.5–5.0s pre-hang window, using 46-dimensional cross-signal kernel telemetry and XGBoost to apply reversible scheduling mitigations before user-visible freeze.

0.5s – 5.0s
Predictive Lead
92.5%
Hang Precision
82% fewer hangs
Freeze Reduction
< 0.5% CPU
Daemon Overhead
Operating SystemsKernel TelemetryPatentMachine LearningReliability
Patent Published · Indian Patent Act 1970 →
ML / AI Research

StrataRL

Forensic-grade GRPO infrastructure for Small Language Models, preventing cross-domain catastrophic forgetting via Stratified Advantage Normalization (SAN).

Qwen2.5-3B
Target Model
1.8GB saved
VRAM Opt.
Per-Stratum Z
Advantage Norm
Zero-PRM Regex
Verification
Reinforcement LearningGRPOSmall Language ModelsReasoningPyTorch
Open Source · Research Framework →
Digital Garden Banyan Tree Illustration
++++
The Knowledge Garden
An interactive knowledge graph built from research notes. Synthesizing 20+ nodes across fluid dynamics, kernel architecture, and agentic security.
v2.0.4-stable© Atharva Mendhulkar 2026
www.mendhu.tech