+ AI engineer from Addis Ababa
+ Building autonomous agents that ship real code to real maintainers
+ 25 PRs to upstream open source. 1 merged into google/garf
+ Knowledge graphs + LLM agents + real-world artifact production
+ 4 GB RAM. Cloud compute. Receipts on GitHub.
- waiting for permission
A 10-tier autonomous research engine that ingests PubMed, grounds entities in PubChem / MONDO / UniProt / MeSH, and runs a 5-expert multi-agent debate to surface calibrated hypotheses.
WIZ-0 → PubMed ingestion
WIZ-1 → Evidence evaluation
WIZ-2 → MeSH / UMLS ontology mapping
WIZ-3 → Consensus grading
WIZ-4 → Contradiction auditing
WIZ-5 → Transitive path discovery
WIZ-6 → 4-pillar validation
WIZ-7 → Researcher prioritization
WIZ-8 → 5-expert debate (Advocate · Skeptic · Methodologist · Translational · Judge)
WIZ-9 → RCT protocol + NIH grant generation
WIZ-10 → self-directed research loop
Stack — Python · httpx · Neo4j · async pipelines · multi-agent orchestration · regulatory grounding
An agent that finds, fixes, and PRs CWE-tagged vulnerabilities across open source. 25 PRs opened across 8 upstream repositories. 1 merged into Google.
Targets hit
| Repository | PRs | CWE classes covered |
|---|---|---|
google/garf |
5 | B307 · B104 · B602 · B324 · ✅ merged |
google/python-fire |
1 | CWE-78 |
microsoft/markitdown |
1 | CWE-611 |
microsoft/agent-governance-toolkit |
1 | CWE-502 |
opencart/opencart |
4 | CWE-941 · security headers |
dograh-hq/dograh |
5 | CWE-78 · CWE-95 · CWE-324 · JWT · SSL bypass |
D4Vinci/Scrapling |
2 | CWE-605 · CWE-502 |
nicobailon/pi-subagents |
1 | CWE-78 |
Stack — Python · static analysis · Bandit · CWE taxonomy · autonomous fork → scan → fix → PR
@@ libs/core/garf/core/cache.py @@
- hashlib.md5(json.dumps(kwargs).encode('utf-8')).hexdigest()
+ hashlib.md5(json.dumps(kwargs).encode('utf-8'), usedforsecurity=False).hexdigest()A 2-character fix that made a Google codebase FIPS-compliant.
status: open to remote AI engineering roles
location: Addis Ababa, ET
focus:
- knowledge-graph + LLM agent systems
- autonomous security tooling
- multi-agent debate / epistemic systems
shipping: WIZ (clinical research) · nanobot (security) · open source PRs
looking_for: remote roles where multi-agent systems are the work
reach_out: github.com/yabets4Constraints are not disqualifications. They are forcing functions.
The receipt is the work, not the resume.
An agent that ships is worth more than an agent that just talks.
The 4GB RAM is the studio. The merged PR is the gallery.
| PRs opened | 25 |
| Upstream repos | 8 |
| Merged at Google | ✅ |
| CWE classes covered | CWE-78 · 95 · 502 · 605 · 611 · 941 · B104 · B307 · B324 · B602 |
| Public repos | 18 |
| Years on GitHub | ~5 (since 2021) |
| RAM | 4 GB |
| Merged PRs ratio | the work speaks for itself |
⟢ the receipt is in the commits ⟢