Skip to content
View MuhammadAashirAslam's full-sized avatar

Highlights

  • Pro

Block or report MuhammadAashirAslam

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.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. 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

Muhammad Aashir Aslam

I'm a software engineering student at NUST in Islamabad. The thing I'm most focused on right now is a research paper on how reliably LLMs call tools when the tool's schema is worded differently: if you rename a tool, rename its parameters, or reword its description, does the model still do the right thing? More broadly I'm interested in AI for software engineering, agentic systems, and making agents easier to observe and debug. I'm also a GSoC 2026 contributor with preCICE.

Website | LinkedIn | Email

Projects and Research Work

  • Tool-calling robustness (paper in progress) : do LLMs still pick the right tool when its name, parameter names, or description are worded differently? We built a pipeline that rewrites tool schemas, translates the model's call back to the original names, and scores it with the benchmark's own evaluators (BFCL and RoTBench). The pipeline passed a 24-case pilot (72 variants), and the full model experiments are next.
  • Argus (TypeScript, MCP, LLMs, Docker): a multi-agent system for investigating and maintaining repositories. Specialized agents look at source code, Git history, CI runs, pull requests, and configuration state, then propose patches. Each patch is verified in a Docker-based check before it is accepted.
  • Kairo (Kotlin, Jetpack Compose, llama.cpp): a privacy-first Android assistant that runs local GGUF models on the phone. It does hybrid on-device retrieval (BM25, dense embeddings, RRF, MMR) and reads documents with OCR. SmolLM2-360M-Q4 ran at 12 to 18 tokens/s in about 270 MB of RAM on ARM64.
  • ConfigLens (Python, in development): a CLI that scans GitHub Actions, Dockerfiles, Compose files, Terraform, and Kubernetes YAML for configuration debt. Rule-based, with severity levels, file and line diagnostics, JSON output, and CI integration.
  • AgentTrace (Python and TypeScript, in development): a local-first observability tool for multi-agent workflows. It records agent steps, tool calls, state changes, latency, and token usage, and shows them on a timeline so you can compare runs and find where one went wrong.

Open source

Google Summer of Code 2026: preCICE (precice.github.io)

preCICE is an open-source library for coupled simulations. I moved its website from Jekyll to Hugo and upgraded the frontend from Bootstrap 3 to 5, with no visual changes.

  • Ported 84 content files, 27 includes, and 7 layouts to Hugo's Go templates, and replaced jQuery with CSS and native Bootstrap 5 components
  • Replaced 10 Git submodules with Hugo module mounts across 14 repos, with namespaced static assets so files don't collide
  • Replaced old Ruby gems in the build, which brought builds under 10 seconds
  • Wrote Node scripts for pushing the search index to Algolia, plus GitHub Actions workflows for build and deploy
  • Kept the old URLs working with Hugo aliases and checked them with automated link validation
  • 40+ merged PRs on CI/CD, search, and accessibility, including a GitHub Actions failure that had been broken for about 1.5 years
  • Worked in small, easy-to-review PRs, with weekly calls with my mentors

Full report : https://gist.github.com/MuhammadAashirAslam/a5c5929b962d0a0886a4f050279fcade

RunAnywhere (wally, their CLI for running open models)

  • #194 (merged): wally decisions --json now includes the request ID and HTTP status. Closes #186. It was part of the v0.8.8 release.
  • #164 (merged): wally uninstall now respects --home over RUNANYWHERE_HOME, so confirming an uninstall can't delete models from the wrong store.

Other work

  • Rawaan: session recording, reviewed notes, and an assistant for doctors and therapists (Next.js, Supabase, Clerk, Groq). Built with a team and picked as one of 300 projects, out of 16k, for the BanoQabil hackathon grand finale.
  • SentientSystems: an AI automation agency I co-founded. We build chatbots, voice bots, agentic workflows, and web apps. One of them is a real-time pixel platform with a 1,000x1,000 canvas and 100k+ monthly users (Redis and WebSocket sync, Stripe, AWS).
  • NUST MCS DEVCON Hackathon: 1st place among 20+ teams, with an AI-powered event manager.
  • Mentoring: ran seminars and bootcamps for 100+ students on contributing to open source, Git workflows, and using AI tools well.

Stack

  • AI: LLMs, RAG, AI agents, MCP, BM25, dense retrieval, GGUF and quantization
  • Languages: Python, TypeScript/JavaScript, Kotlin, Java, C/C++
  • Frameworks: React, Next.js, Node.js, NestJS, Jetpack Compose, Hugo, LangChain, LlamaIndex, llama.cpp
  • Data: PostgreSQL, MongoDB, MySQL, Redis, Supabase
  • Infra: AWS, Docker, Git, GitHub Actions

Pinned Loading

  1. Argus Argus Public

    ARGUS is an evidence-driven multi-agent system for repository analysis and autonomous maintenance. It investigates codebases, detects configuration debt, reasons over repository history, generates …

    TypeScript 1

  2. kairo kairo Public

    Kairo is an open-source, modular, and privacy-first Android application designed for local Retrieval-Augmented Generation (RAG), clinical consultations, document synthesis, and conversational coding.

    Kotlin 1

  3. ConfigLens ConfigLens Public

    ConfigLens is an open-source CLI that scans a repository's DevOps configuration files — GitHub Actions workflows, Dockerfiles, Docker Compose files, Terraform, and Kubernetes manifests — and detect…

    Python 1

  4. AgentTrace AgentTrace Public

    A lightweight, local-first observability tool for multi-agent workflows. Trace agent steps, tool calls, errors, and handoffs to pinpoint failure evidence, inspect timelines, and compare runs with z…

    Python 1