Varun Pratap Bhardwaj — AI Reliability Engineering

I build Qualixar, my independent research initiative for AI agent reliability: decision routing, governed memory, behavioral contracts, statistical testing and verified execution.

10 core products · 10 public arXiv preprints

Start with the task you need to solve. Each product below links to its source, practical setup and supporting evidence.

Qualixar · My work and writing · SuperLocalMemory · AgentAssert

All 10 products

Choose the capability your workflow needs, then follow the setup, examples or research in Resources. Product names link directly to the source repositories.

Product	Description	Focus	When to use it	Resources
Jev Decision Layer	Returns advisory choices from explicit options, with reasons, constraints and decision receipts.	Decision routing	Choose a task, tool, model tier or review route.	Product Recipes Guide
Bounded Loops	Runs agent loops and graphs within declared limits; independent gates check completion and hash-chained ledgers record results.	Verified execution	Fix failing tests or repeat a workflow until its acceptance check passes.	Product Recipes Proof
SuperLocalMemory	A governed memory control plane combining persistent context, multi-channel retrieval, knowledge graphs and workspace isolation.	Memory & governance	Carry project knowledge across agent sessions while controlling access, retention and context.	Product Workflows Research
SLM MCP Hub	Federates MCP servers behind three discovery-and-call tools, with shared backends and on-demand tool lookup.	Tool federation	Connect multiple clients to many servers without loading every tool definition up front.	Product Setup Routing
SkillFortify	Scans agent skills and MCP configurations using capability analysis, dependency checks and supply-chain reports.	Skill security	Inspect skills before installation and add security checks to your delivery pipeline.	Product Guide Synthetic benchmark
AgentAssert	Defines behavioral contracts and enforces hard or soft constraints against application-supplied agent state.	Runtime contracts	Check permissions, limits and required signals at an action boundary.	Product Templates Guide
AgentAssay	Measures behavioral regressions with execution traces, statistical verdicts and adaptive trial budgets.	Testing & measurement	Compare agent behavior after changing prompts, models or tools; reuse existing traces.	Product Guide Paper
Agent Amplifier	Adds effort routing, goal anchoring, convergence checks and token budgets to coding-agent sessions.	Coding runtime	Keep coding sessions focused and manage effort through supported host adapters.	Product Setup Post
SLM Mesh	Connects agents across harnesses with peer messages, shared state and advisory file locks, locally or across a LAN.	Agent coordination	Coordinate work between agent sessions on one computer or multiple networked machines.	Product Setup Post
Qualixar OS	Designs and runs agent teams, evaluates outputs through a judge pipeline, and retries with structured feedback.	Orchestration & judging	Run multi-agent tasks with quality review, model routing and budget tracking.	Product Setup Post

Use each repository’s setup guide for supported hosts and operating modes. Benchmark results apply to their published datasets and protocols; the SkillFortify benchmark linked above uses generated specimens.

Research: 10 public arXiv preprints

These public arXiv records are preprints. Read each paper for its assumptions, version, experimental scope and limitations. Listing a paper beside a product does not establish that every current feature was evaluated in that paper.

Paper	arXiv	Related project
Bounded Loops: Pre-Run Spend Bounds, Proved Termination, and Verified Completion for Agent Harnesses	2609.27871	Bounded Loops
Agent Behavioral Contracts II: Certifying Compositional Reliability Without Assuming Independence	2608.12895	AgentAssert
SuperLocalMemory 4.0: The Governed Memory Operating System for AI Agents	2608.08253	SuperLocalMemory
Qualixar OS: A Universal Operating System for AI Agent Orchestration	2604.06392	Qualixar OS
SuperLocalMemory V3.3: The Living Brain -- Biologically-Inspired Forgetting, Cognitive Quantization, and Multi-Channel Retrieval for Zero-LLM Agent Memory Systems	2604.04514	SuperLocalMemory
SuperLocalMemory V3: Information-Geometric Foundations for Zero-LLM Enterprise Agent Memory	2603.14588	SuperLocalMemory
AgentAssay: Token-Efficient Regression Testing for Non-Deterministic AI Agent Workflows	2603.02601	AgentAssay
Formal Analysis and Supply Chain Security for Agentic AI Skills	2603.00195	SkillFortify
Agent Behavioral Contracts: Formal Specification and Runtime Enforcement for Reliable Autonomous AI Agents	2602.22302	AgentAssert
SuperLocalMemory: Privacy-Preserving Multi-Agent Memory with Bayesian Trust Defense Against Memory Poisoning	2603.02240	SuperLocalMemory

Build with the tools

Try a recipe, inspect the source, and share the result in Qualixar Discussions. Ask setup questions in Q&A, propose a workflow in Ideas, or report a reproducible bug in the relevant repository.

If a product helps your work, star its repository to follow it. Access and participation remain open without a star.