Collins Karani

Multi-disciplinary technology professional combining certified cybersecurity expertise, production-grade AI systems engineering, advanced data annotation proficiency, and rare native trilingual ability (Gikuyu, Swahili, English). Registered as M'Munya Collins Karani at Maseno University's School of Computing and Informatics (CCS/00052/024), with formal cybersecurity credentials from Emobilis Institute of Technology (2023–2025) and 2+ years of active bug bounty research on HackerOne. Engineered HERMAN — a production AI customer support system deployed for a real Nairobi retail business , an autonomous LLM evaluation agent built on the Anthropic API.

Independently benchmarked and evaluated outputs from 5+ large language models, producing structured annotation reports across accuracy, fluency, cultural relevance, and safety dimensions. Fluent in OSS-fuzz vulnerability research methodology, Power Query / VBA Excel automation, structured linguistic data annotation, and async independent delivery at professional standards. Available immediately, fully remote-ready, and equipped to contribute expert-level output across cybersecurity, AI training data, spreadsheet modeling, and multilingual language tasks.

Vulnerability Researcher — Bug Bounty Hunter at HackerOne Platform (2023-01 – Present)

• Spearheaded independent vulnerability research across multiple live production targets, identifying, reproducing, and responsibly disclosing security flaws with structured proof-of-concept reports, step-by-step reproduction instructions, and validated remediation patches — executing the identical find-document-patch loop that OSS-fuzz-style AI training data tasks require.
• Executed deep low-level code analysis in C and C++ to surface memory safety vulnerabilities including buffer overflows, use-after-free conditions, heap corruption, and integer overflow issues — core competencies directly applicable to generating and patching OSS-fuzz-style code repositories.
• Delivered 100% first-submission acceptance rate on all disclosed findings by applying rigorous documentation standards, precise impact assessments, and reproducible proof-of-concept construction across every vulnerability report submitted to the platform.
• Operated fully asynchronously across evening and weekend research sessions with zero supervision, maintaining consistent output quality and meeting platform submission standards independently — demonstrating the self-directed async delivery discipline remote project-based roles demand.
• Leveraged Kali Linux (WSL2), Burp Suite, GDB, Valgrind, Nmap, and Wireshark as a daily security toolkit, building professional-grade offensive security execution instincts across web, API, and binary attack surfaces.

Lead Developer & AI Systems Engineer — HERMAN Project at Urban Hype KE / Maseno University (2024-01 – Present)

• Architected and engineered a full-stack production AI customer support chatbot in Java with a MySQL backend and HTML/CSS/JavaScript frontend, deployed for an active Nairobi fashion retail business — delivering a real-world AI system from concept to production within a five-person team.
• Accelerated AI response quality by 40% across iterative annotation cycles by systematically evaluating, scoring, and ranking 100+ model outputs against structured accuracy, tone, safety, and cultural relevance rubrics — building structured data annotation instincts applicable across AI training workflows.
• Hardened the production backend against 6 identified exploitable attack surfaces including SQL injection vectors, privilege escalation risks, and authentication bypass vulnerabilities by conducting a full pre-deployment security audit and producing documented patches with root cause analysis and regression test cases for each finding.
• Engineered a Power Query and VBA-driven operational Excel dashboard for management reporting, reducing manual reporting time by 60% by consolidating disconnected spreadsheets into a single automated model with dynamic pivot tables, KPI visualisations, and reproducible refresh logic.
• Produced comprehensive structured technical documentation spanning DFDs, UML diagrams, ER diagrams, system architecture flowcharts, and written methodology reports — achieving zero revision requests from project stakeholders on all final documentation deliverables.

Independent AI Model Evaluator & Benchmark Analyst at Remote (2024-01 – Present)

• Evaluated and benchmarked outputs from 5+ large language models — including Gemma 4, LLaMA variants, and the GLM ecosystem — across accuracy, fluency, factual consistency, cultural fit, and safety dimensions, producing structured written assessment reports across 10+ benchmark sessions.
• Constructed OMNISCIENT, an autonomous React + Anthropic API knowledge evaluation agent, to systematically search, annotate, and critique AI-generated content quality — demonstrating end-to-end familiarity with AI data collection pipelines and training data quality standards.
• Identified and documented edge cases, hallucination patterns, and out-of-distribution failure modes across multiple LLM evaluation sessions, producing structured written justifications and escalation documentation that directly mirrors professional data annotation workflows.
• Deployed and managed local LLM inference environments using LM Studio and Ollama on SAPIENTIA (Lenovo ThinkPad X1 Yoga, Windows 11 Pro), integrating Gemma 4 E4B for offline evaluation alongside cloud-based API workflows — demonstrating both local and API-side AI toolchain fluency.

Excel Modeling Specialist & Python Automation Developer at Independent / Academic (2022-01 – Present)

• Designed and delivered multi-sheet operational Excel models featuring advanced formulas (INDEX/MATCH, XLOOKUP, dynamic arrays, LAMBDA), Power Query M-language transformation pipelines, and pivot table dashboards — reducing manual reporting overhead by 60% on the HERMAN project's management reporting layer.
• Engineered a VBA macro suite automating 8 repetitive data entry and report-generation workflows, cutting weekly administrative overhead from 4+ hours to under 30 minutes while eliminating transcription errors across all automated processes.
• Developed an openpyxl + pandas Python pipeline to automate generation of structured Excel workbooks from raw JSON API responses, reducing manual data formatting from 2+ hours per batch to under 3 minutes while preserving full conditional formatting and formula integrity.
• Tutored 3 university peers in advanced Excel techniques — VLOOKUP-to-XLOOKUP migration, Power Query fundamentals, and pivot table design — reinforcing a teaching and communication ability applicable to structured annotation guideline interpretation and knowledge transfer tasks.

Gikuyu Language Specialist & Linguistic Data Contributor at Nairobi, Kenya (Present)

Native Speaker · Lifelong

• Translated, reviewed, and edited content between English and Gikuyu as a lifelong native speaker (Karani lineage, Central Kenya), maintaining grammatical accuracy, cultural relevance, tonal consistency, and natural idiomatic expression across all linguistic outputs.
• Applied native Gikuyu domain expertise to evaluate AI-generated language outputs for cultural authenticity, dialectal accuracy, and contextual appropriateness — providing the rare human-expert linguistic judgment that AI language training data pipelines require.
• Demonstrated fluency across East African cultural contexts in English, Gikuyu (C2), and Swahili (C2), enabling high-fidelity localisation, back-translation, and culturally-grounded linguistic annotation across all three languages simultaneously.

Diploma in Cybersecurity in Cybersecurity – Emobilis Institute of Technology, Westlands, Nairobi (2023-01 – 2025-12)

B.Sc. Computer Science & Informatics in Computer Science & Informatics – Maseno University, School of Computing and Informatics (2022-01)