first-person nerd | bits & bytes tinkerer

recent focus: analyzing prompts for intent-tribes, making observability and telemetry actually testable (so your LLM pipelines don't fly blind), giving financial research some structure and compliance muscle, and feeding AI agents APIs that are fast and cheap. if it has an algorithm and a purpose, i'm probably into it.

old habit / lasting interest: signal processing and knowledge representation. spread-spectrum watermarking in MATLAB, ConceptNet clients and graph walks in Python/CoffeeScript, AODV on Zigduino, Booth's multiplier in MIPS — that era still shapes how i think.

what's on the bench right now (repos)

• 🧠 semantic-clustering — "turn prompt vibes into points, then let density find the tribes" analysis.
• 📡 telemetry-simulator — fake traces, metrics & logs so convincing your observability stack will write home about them. agentic planning, tool recs for LLMs, tool calls with outer + inner policy-controller traces (yes, both layers)—the full agent stack, simulated. schema-driven OpenTelemetry for LLM land. "it works on my machine" should include "and my telemetry too."
• 📈 copinance-os — Copinance OS: fin research with structure. when "just use a spreadsheet" isn't enough. compliance and "yes, we actually checked" energy.
• 🤖 onemcp — API access for AI agents: fast, accurate, cost-efficient. (OneMCP) agents gotta eat too.

(reminder to myself: update this whenever the backlog wins or a new obsession hits.)

from the signal-processing & knowledge-representation era — still proud of these

• 🎭 cox-watermarking — hide bits inside other bits. spread-spectrum informed watermarking in MATLAB. 29 ⭐, 15 forks. steganography's nerdy cousin.
• 🕸️ conceptnet5-client — walk the knowledge graph. reason over it. Python client + graph-walk inference for ConceptNet5. coworked with MIT Media Lab — good old graduate school days...
• 📶 zigduino-aodv-routing — AODV on Zigduino. mesh routing in C. packets, uh, find a way.
• 🔗 semantic-relation-composition — reason over relations. graph-walk style. semantic composition go brr.
• 😤 sentiment-analyzer — Chinese sentiment orientation for discourse relation pairs. ActionScript. feelings, quantified.
• ⛓️ forward-chaining-reasoning — naïve forward chaining in Lisp. if A and B then C. the classic.
• 👁️ Visualize-ConceptNet — the graph, visualized (CoffeeScript). see it.
• 🏷️ concept-tag-generator — content → concepts → tags, powered by ConceptNet. NLP without the buzzword bingo.
• 📝 csie-parser — parser for an "imaginative" csie programming language. because every language deserves a parser.
• 🔍 sourcecode-scanner — source code scanner written with flex. tokens before they were cool.
• 🧮 booth-multiplicator — Booth's algorithm in MIPS assembly. we multiply the hard way.
• 🎵 music-compiler — compile text-based musical compositions. C. because code and music share a grammar.

• languages: Python, Java, Scala, MATLAB, C, MIPS assembly, CoffeeScript, TypeScript
• domains (now): observability & telemetry, financial research, AI/ML tooling
• domains (old habit): signal processing, knowledge representation, embedded / wireless, computer architecture
• mood: "what if we implemented that paper / that algorithm / that protocol ourselves?"

i'm into open source, clean implementations, and projects that mix theory and practice. if you're building something with observability, telemetry, knowledge graphs, financial research, or weird embedded routing or you just want to nerd out about algorithms — say aloha!

"fake traces, real intent. spreadsheets need backup. agents gotta eat."
— probably me, after too much coffee

_{profile readme · israkir · github.com/israkir · repos}