University student · AI/ML · Automation · INOVENS
ege = {
"role": "AI Engineer & Full-Stack Dev",
"focus": ["LLM Automation", "Self-hosted AI", "Open Source"],
"stack": ["Python", "TypeScript", "Docker", "Linux"],
"affiliation": "INOVENS · Hackathons · Client Projects",
"building": ["mcp-forge", "mcp-sanity"],
"open_source": "42 PRs across 24 upstream repos, September 2026",
"motto": "Automate everything. Ship fast. Keep it simple."
}
Languages
Frontend & Mobile
Backend & Data
AI / LLM
Infra & Tools
42 PRs · 24 upstream repos · 9 merged · 22 in review · September 2026
Merged
| PR | Repo | Change |
|---|---|---|
| #48864 | huggingface/transformers (166k⭐) |
fix(vibevoice-asr): use integer ceiling division for audio token count |
| #554 | agentguard-ai/tealtiger |
docs(mcp): add Continue + Windsurf client configs (follow-up to #548) |
| #303 | sara-czasak/py-simple-wrap |
feat(easy_random): add generate_password helper |
| #961 | MakazhanAlpamys/Soup (6.9k⭐) |
feat(mlx): widen sft table to remaining four CUDA-only fields |
| #957 | MakazhanAlpamys/Soup (6.9k⭐) |
docs(readme): polish Turkish translation for v0.75.0 release |
| #81 | open-devhub/website |
fix(footer): replace All rights reserved with GPL-3.0 license link |
| #629 | yakew7/Fair-Code |
docs(profiler): correct bias audit count from six to seven |
| #30318 | lingdojo/kana-dojo (3.4k⭐) |
content: add new japanese false friend |
| #39 | gczobel/calibre_mcp_server |
docs(tool-surface): remove stale claim that non-link custom columns retu… |
In review — selected open PRs
| PR | Repo | Change |
|---|---|---|
| #18520 | ollama/ollama (181k⭐) |
Allow 96-char model names so long HF GGUF pulls validate |
| #57518 | vllm-project/vllm (92k⭐) |
LoRA device lookup falls through to packed weights when repacking clears… |
| #8248 | microsoft/autogen (61k⭐) |
Carry JSON Schema type for optional tool fields |
| #41572 | BerriAI/litellm (59k⭐) |
fix(proxy): guard SlackAlerting periodic_flush start on repeated update_… |
| #41243 | BerriAI/litellm (59k⭐) |
fix(passthrough): keep adaptive thinking/effort on native /v1/messages p… |
| #41057 | BerriAI/litellm (59k⭐) |
fix(bedrock): keep mid-conversation system messages in place on converse… |
| #41012 | BerriAI/litellm (59k⭐) |
fix(proxy): case-insensitive User-Agent header lookup in _get_user_agent… |
| #13278 | continuedev/continue (36k⭐) |
fix: correlate id-less parallel tool-call fragments by stream index |
| #5133 | PrefectHQ/fastmcp (28k⭐) |
fix(resources): strip URI fragment before matching query params |
| #12799 | deepset-ai/haystack (27k⭐) |
LLMRanker returns documents unranked on None query instead of crashing |
| #2830 | modelcontextprotocol/typescript-sdk (13k⭐) |
Settle pending requests when the SSE response leg ends early |
| #25 | SnailSploit/Claude-Red (6.3k⭐) |
fix: use RFC-reserved names instead of live attacker.com in examples |
Plus open work in sarracenia, zemberek-nlp, whisper-asr-webservice, astronomer-cosmos, petdex, ddgs, memtrace-public.
| Project | What it does |
|---|---|
mcp-forge |
One-command MCP server scaffolder — Python/TS templates, Docker, inspector-verified handshakes |
mcp-sanity |
Live doctor for MCP configs — discovers Claude Desktop/Cursor/Codex servers, spawns each one, probes the handshake, names the broken one |
