01 · Roasts
The Burst Shipper
supplylens: 5 commits in 2 hours. flowlink: 9 commits in 1 day. Shrink-challenge: 2 commits in 5 minutes. You don't build software, you summon it in fever dreams and then move on before the IDE finishes indexing.
84% Jupyter, 0% Regrets
Your language breakdown is literally 84% Jupyter Notebook. That's not a portfolio, that's a very long research paper that forgot to become a product.
CI? Never Heard of Her
Out of 12 analyzed repos, exactly 2 have CI (gpucheck, pixmask). You've shipped a GPU testing framework and an LLM security library but somehow forgot to apply those same standards to 83% of your own work.
License Collector (Void Edition)
adaptconfig claims MIT in the README but has no LICENSE file. supplylens: same. medscribe-ai: same. You keep writing 'MIT License' as if typing it conjures legal protection out of thin air.
EuroLLVM Dropout
You had a paper accepted at EuroLLVM 2026 (libkdl) and it still has 0 stars and no tests. You presented at a compiler conference and your own repo didn't notice.
Built using
Zoral
Shadows one worker for a week, then takes over their job with zero extra setup. Behaves exactly like the original.
zoral.ai
02 · Category breakdown
- Impact25% weight68C
- Consistency20% weight65C
- Quality20% weight69C
- Depth15% weight58D
- Breadth10% weight65C
- Community10% weight55D
03 · Stats
365-day commit heatmap
127 active days
Language distribution
- Jupyter Notebook84%
- Python8%
- TypeScript2%
- JavaScript2%
- C1%
- HTML1%
- Other2%
04 · Numbers
Owned repos
non-fork
70
Commits
last 12 months
721
Followers
41
Joined GitHub
Oct 2022
05 · Top repos
Akasxh /
adaptconfig
Hackathon project: AI-powered API integration config platform with FastAPI backend, React frontend, document parsing, LLM-driven field mapping, and simulation testing. Shipped live on Railway with 899 passing tests, comprehensive docs, and multi-adapter support for Indian fintech.
Akasxh /
libkdl
GPU kernel runtime dispatch system with 92MB codebase (5,100 LOC libkdl + benchmarks), presented at EuroLLVM 2026, adding vendor-agnostic OffloadBinary selection via MLIR metadata vocabulary and per-layer latency analysis. Strong research foundation but no tests/CI and pre-conference status limits adoption signal.
Akasxh /
gpucheck
Specialized pytest plugin for GPU kernel testing with dtype-aware assertions, parametric testing decorators, CUDA benchmarking, and fuzzing. 3 days old, foundational architecture but 1 star, no external adoption yet.
Akasxh /
pixmask
Specialized image sanitization library for multimodal LLM security. Pure C++ core with SIMD (Highway), Python bindings via nanobind. v0.1.0 ships 5-stage pipeline (validate, decode, bit-depth reduction, median filter, JPEG roundtrip). 527KB codebase with comprehensive architecture docs, CI/CD on 3 platforms, but early-
Akasxh /
flowlink
TypeScript agent-native payment layer on HashKey Chain, freshly architected with modular compliance-first design. Early stage (9 commits in ~1 day), 0 stars, incomplete implementation (Layers 2–9 in progress). Strong skeletal structure with independent lib modules, comprehensive error handling, ed25519 receipt signing,
Akasxh /
claude-forge
Self-evolving multi-agent workforce for Claude Code with research/engineering/forge teams. Typed Python + substantial documentation (PROTOCOL.md files, agent personas), structured architecture, but minimal community adoption (3 stars, 0 forks) and only 3 days old.
Akasxh /
medscribe-ai
Hackathon-stage clinical scribe PWA with Gemini-powered FHIR R4 extraction, real-time CDS alerts, and Hindi-English support. Early-stage but architecturally ambitious; ships typed Python backend, structured React frontend, and comprehensive domain logic (15+ drug interactions, 8 FHIR resource types, specialty-aware ext
Akasxh /
llm-pulse
Personal project tracking LLM benchmarks via automated weekly briefs. Has README, CI/CD pipeline, and structured architecture (fetch → analyze → dashboard), but no tests, no license, untyped HTML/JS, and minimal adoption signals (0 stars, 7 of last 30 commits).
Akasxh /
supplylens
Early-stage supply chain intelligence SaaS with FastAPI backend + vanilla JS frontend. Integrates Crustdata, Gemini, and Sarvam APIs for supplier discovery, legitimacy scoring, and personalized outreach. Clean typed Python, structured project, but nascent (0 stars, 5 commits, created Apr 2026).
Akasxh /
Shrink-challenge
Experimental challenge entry for model compression: ultra-tiny student networks (~13K params) distilled from WideResNet teacher with novel 4/2-bit codebook quantization. Typed Python, structured layout, clear docs, but minimal community adoption (0 stars), very recent (2 days old), single commit burst.
Akasxh /
Akasxh.github.io
Personal GitHub Pages repository with 10 MB of HTML/assets. 30 recent commits over ~2 years but no documentation, tests, CI, or README. Appears to be a portfolio/personal site with no external adoption or influence.
Akasxh /
flowlink-ppt
Archived repository redirecting to main flowlink repo. README explicitly states this is deprecated; all content moved. 3 commits in ~1.5 hours, 0 stars, HTML-only staging area for pitch slides.
06 · Timeline
- Oct 11, 2022Joined GitHub
- Dec 23, 2023Created Akasxh.github.io
- Oct 26, 2025Created pixmask
- Mar 18, 2026Created medscribe-ai — Speak naturally with your patient. Get structured clinical notes
- Mar 22, 2026Created llm-pulse — Weekly AI model intelligence brief — tracks LLM performance, rankings, and benchmark trends
- Mar 24, 2026Created gpucheck — pytest for GPU kernels — correctness, benchmarking, and fuzzing for CUDA and Triton
- Mar 26, 2026Created adaptconfig — AI-Powered Integration Configuration Platform | Team Nucleolus | FinSpark Hackathon IIT Patna
- Mar 29, 2026Created Shrink-challenge
- Apr 10, 2026Created libkdl — Kernel Dynamic Linker — ld.so for GPU kernels. Runtime vendor-agnostic dispatch of ML kernel binaries via MLIR OffloadBinary. EuroLLVM Dublin 2026.
- Apr 13, 2026Created claude-forge — Self-evolving multi-agent workforce for Claude Code. Research teams that investigate, engineering teams that ship, a capability forge that builds new skills, and a memory layer tha
- Apr 19, 2026Created supplylens — Supply chain intelligence tool — find suppliers, verify legitimacy, generate personalized outreach. Powered by Crustdata + Gemini + Sarvam AI.
- Apr 22, 2026Created flowlink — Agent-native compliance-first payment layer on HashKey Chain. Markdown is the API.
- Apr 22, 2026Created flowlink-ppt — FlowLink pitch slides + agent-native comparison (real-run logs)
- Apr 23, 2026Most recent push to flowlink
07 · Compare
08 · Rubric
How this score was produced
Overall = Σ (category × weight) + gentle top-end curve
Tier thresholds
▸ How the pipeline works
- 01Scrape.Pull every non-fork repo pushed in the last 90 days, plus your contribution calendar, followers, and language byte counts — straight from GitHub's REST & GraphQL APIs.
- 02Triage.A small model reads every repo's file tree + README and picks the 20 files per repo that actually reveal how you code.
- 03Grade each repo. All repos run in parallel through a fast scoring model that reads the picked files and rates each one independently on Impact, Quality, and Depth — with evidence citations.
- 04Aggregate. A larger reasoning model combines the per-repo scores with server-computed stats (heatmap, commit cadence, language entropy, follower count) to produce the 6-dimension profile score + roasts.
- 05Correct.Deterministic server-side checks enforce anchor-scale floors (e.g. a profile with 2,000+ public commits can't score 30 Consistency) and recompute the final verdict.
~90 seconds per profile, ~$0.25 in compute. Total of ~240 files read across your top-12 repos. One rating per GitHub account per day.
▸ Data sources & caveats
- Heatmap & commit totals: GitHub GraphQL
contributionsCollection— covers the last 365 days, includes private repos when the user has opted in (default). - Language %: byte totals across the top 30 owned non-fork repos.
- Curve: a small upward nudge centered on raw score ≈ 70, capping at 100. Prevents specialists from being unfairly penalised for narrow breadth.
- Anchor corrections: when server-measured signals (e.g. privateWorkLikely, multiRepoVolume, follower count) mandate a minimum category score, the aggregation step enforces it. These are signal-conditional, not identity-based floors.