Skip to content
View desenyon's full-sized avatar
11 followers · 7 following

Achievements

Achievement: Pair Extraordinairex3Achievement: Pull Sharkx2Achievement: YOLOAchievement: Quickdraw

Achievements

Achievement: Pair Extraordinairex3Achievement: Pull Sharkx2Achievement: YOLOAchievement: Quickdraw

Highlights

Block or report desenyon

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
desenyon/README.md 

 ███╗   ██╗ █████╗ ██╗████████╗██╗██╗  ██╗
 ████╗  ██║██╔══██╗██║╚══██╔══╝██║██║ ██╔╝
 ██╔██╗ ██║███████║██║   ██║   ██║█████╔╝ 
 ██║╚██╗██║██╔══██║██║   ██║   ██║██╔═██╗ 
 ██║ ╚████║██║  ██║██║   ██║   ██║██║  ██╗
 ╚═╝  ╚═══╝╚═╝  ╚═╝╚═╝   ╚═╝  ╚═╝╚═╝  ╚═╝

founder @ Concatenate



desenyon · naitik gupta · california



Portfolio LinkedIn Research ORCID


◈   what i build   ◈


I write code that is either something I wanted to try because it sounded cool,
infrastructure I was tired of not having, or both.


Most of my work lives at Concatenate — the compile step between scattered company knowledge
and the agents that need to act on it. I also ship quant research tools and applied ML
when the problem is interesting enough to justify the math.


❖   concatenate   ·   agent infrastructure   ❖


concatenate-app
context runtime

The compile step for enterprise agents. Scattered knowledge — repos, Slack, specs, policies — becomes typed, permissioned, source-backed Context Packets before an agent acts.

collective
organizational memory

Infrastructure layer for AI software engineering. Codex, Claude Code, Cursor, and Copilot agents reuse verified context instead of rediscovering the same repository every session.

hypercode
intelligence layer

HyperGraph, HyperContext, HyperMemory — and a beautiful TUI — for coding agents.

creation
agent OS

Local-first operating system for builders who want autonomous software delivery without stitching together a dozen third-party APIs.

collective-compress
context compression

Learned context compression for LLMs — part of the Collective stack.

 


◆   developer tooling   ◆


converge
Proves and repairs dependency topologies. Finds the cycle, proves it, fixes it.

infinitecontex
Local-first project memory — compact context so any agent understands your repo immediately.

dex
Developer cockpit in one Go binary — network, API, JSON, regex, files, clipboard, and more.

thunder
tn — ripgrep + skim + thefuck in two keystrokes.

ClaudeTube
YouTube in your Cursor sidebar — watch, queue, agent-control while coding.

+ a lot more on the repos tab →


◈   quantitative finance   ◈


ephemeral
research desk

Keyboard-first terminal workstation — watchlists, news, backtests, LLM routing, artifacts. Markets, models, and decision loops in one shell.

flux-rx
Publication-quality interactive charts and dashboards for any stock, ETF, or index.

sentinel
Safety-first equity stat-arb — risk-first, ordinal-only, signal lifecycle.

quant research
Systematic strategies on QuantConnect. Desenyon Trade Group — #2 Strategies Leaderboard.


❖   research   ·   evals   ❖


GRAFT-Net
Graph-routed adaptive fusion transformers.

ampp
Autonomous theorem proving with machine-checked guarantees.

aurane
ML DSL → idiomatic Python / PyTorch.

updraft-lm
117M-param transformer from scratch — the math, not just .from_pretrained().

scope-rx
Attribution, metrics, and visualization for interpretability.

alcas-a1
Antibody sequence landscapes — Synopsys 1st, Comp. Biology.

pressbench
Measuring sycophancy — do models hold correct beliefs under social pressure?

lucenteval
Adversarial eval platform for AI agents — 6 dimensions, public leaderboard.


◆   credentials   ◆
◇ Engineer Intern · DataRobot — enterprise AI platform
◇ Agentic AI Intern · RagaAI — clinical-trial screening agent, modular RAG at scale
◇ #2 · QuantConnect Strategies Leaderboard
◇ 1st Place · Synopsys Science Fair — Computational Biology
◇ Qualifier · California State Science & Engineering Fair
◇ Published · Journal of Student Research
◇ 15+ articles · RandomResearchAI — 20k+ reads


◈   favorite technologies   ◈


Python Rust Go TypeScript React Next.js


PyTorch Docker Postgres Git Linux Vercel


❖   github   ❖


github stats

  

top languages



streak stats



cooked

Pinned Loading

  1. Fireguard Fireguard Public

    Congressional App Challenge - FireGuard - Special Recognition

    Dart

  2. infinitecontex infinitecontex Public

    Local-first project memory for AI coding workflows.

    Python

  3. ampp ampp Public

    AMPP is an autonomous system that proves mathematical theorems with machine-checked guarantees. It targets advanced combinatorics, number theory, and Erdos-style problems — the kind of mathematics …

    Python

  4. converge converge Public

    Mathematically proves and automatically repairs your dependency topologies.

    Python

  5. GRAFT-Net GRAFT-Net Public

    Graph-Routed Adaptive Fusion Transformer Network — GRAFT-Net v0.1.0

    Python

  6. pressbench pressbench Public

    Pushback Resistance & Epistemic Stability Score - A standardized framework for quantifying sycophancy — measuring how confidently language models hold correct beliefs under social pressure.

    Python