I build lightweight tools for Windows, AI coding workflows, and low-level runtimes in C++ and Python. Most of what I make starts because an existing tool was too slow, ate too much memory, or cost $50/month for simple API calls.

Email • LinkedIn • GitHub

A bezel-flush status HUD that doesn't steal focus.

• Background Electron widgets eat 200MB+ of RAM just to display a few status badges.
• Built in native C#, WPF, and PowerShell with global hotkeys (Shift+Enter) to confirm background tasks without task switching.
• 0.0% idle CPU, <25MB RAM • C#, Win32, WPF, PowerShell • Apache 2.0

Give coding agents an instant structural map instead of grepping thousands of lines.

• AI coding assistants burn through context windows and generate hallucinated imports when reading raw files blindly.
• An MCP server that parses code into deterministic AST symbol graphs with SQLite WAL caching.
• <10ms symbol lookup, cuts agent context token burn by >90% • Python 3.10+, AST, MCP Protocol, SQLite • Apache 2.0

Catch non-progressing agent loops before they drain API credits.

• Autonomous agents often get trapped in retry cycles, repeating failing tool calls until context runs out.
• Hashes actions and models execution history as a directed graph to detect back-edges and cycles in real time.
• <1ms cycle fingerprinting, saves up to 42% of token budgets • Python 3.10+, DAG Trajectory Engine • Apache 2.0

Wipe sensitive keys and credentials from memory without compilers optimizing the wipe away.

• Standard free(), delete, or memset() calls are frequently stripped by modern compilers via Dead Store Elimination.
• Header-only C++20 library and Python module enforcing hardware memory barriers and 3-pass DoD 5220.22-M overwrites.
• 0.08 µs cache-line purge, zero forensic plaintext leakage • C++20 Header-Only, Python ctypes • Apache 2.0

Send links and automate DM leads without third-party SaaS subscriptions.

• Commercial tools like ManyChat cost $50+/month and require full cloud access tokens just to send automated links.
• Zero-dependency Python CLI and direct API agent with sub-16ms dispatch and an interactive terminal simulator.
• <16ms message dispatch, <25MB RAM, $0 SaaS fees • Python 3.10+, Direct Web API • Apache 2.0

Prime AI coding agents with your actual working context instead of starting from scratch.

• Task lists fall out of date within 48 hours, and coding agents have zero idea what you were just working on or what broke.
• Merges active window pulses with recent session log mining to produce an instant context brief.
• <0.5ms pulse merge, streams session logs in <12ms, emits a <80 token prompt prime • Python stdlib, Win32 ctypes • Apache 2.0

Catch leaked credentials and SQL injections without waiting minutes for heavy SAST containers.

• SAST tools like Bandit and Semgrep require heavy setups, take 3–8 seconds, and regex scanners drown in false positives.
• Pure Python standard library scanner with Shannon entropy testing ($H \ge 3.2$) to verify cryptographic randomness and AST grammar trees to isolate real SQL injections.
• ~0.15s scan per 10k LOC, 0 dependencies • Python 3.9+, AST, stdlib • Apache 2.0

• Standard library first: Zero third-party dependencies when language primitives do the job.
• Low memory ceilings: Keep background processes under 25MB RAM.
• Local and private: No sending local context or credentials to unnecessary cloud backends.
• Real benchmarks over marketing: Measure latencies in microseconds and milliseconds, not vague adjectives.

• Email: jaswanthreddy1537@gmail.com
• LinkedIn: linkedin.com/in/kannali-sai-jaswanth-reddy-aa3678337
• GitHub: github.com/Jaswanth1902