Kyrylo Kholodenko

Problem

A few hundred bots scan marketplaces, social networks and piracy sites millions of times a day, and every site keeps adding anti-bot defences.

What I did

• Kept bots reliable against Cloudflare, Akamai, proof-of-work challenges, TLS fingerprinting, geo-blocking and captchas: diagnosis first, then the cheapest fix that passes.
• Fixed the shared framework, not single bots: a proof-of-work solver that unblocked a whole family of mirror sites, and a page-ownership contract for every Playwright bot.
• Shipped data-acquisition features: search by image, delivery-location-aware monitoring that never reports a false “closed”, signed-JWT auth for a mobile API.

Problem

Open edX operators had no built-in way to issue recognised digital badges or verifiable credentials.

What I did

• Built the Credly integration end to end: new platform events, an event-bus consumer and a badges service in Credentials.
• Added Accredible as a second provider through a provider-agnostic client layer.
• Implemented W3C Verifiable Credentials / Open Badges v3 issuance with revocation status lists.

Problem

Teachers needed development programmes matched to their assessed competencies, not a generic catalogue.

What I did

• Built semantic matching on OpenAI embeddings in pgvector, kept fresh by Celery bulk jobs and on-save refresh.
• Shipped LLM-written assessment conclusions, then moved them to a smaller model to cut inference cost.
• Ran the stream as main reviewer and release manager.

Problem

Teachers and programme owners wanted answers from learner activity (who is stuck, which course rates low) without asking engineers for a report.

What I did

• Rolled out and extended the Open edX analytics stack (tracking events as xAPI, ClickHouse, dbt, Superset) on five client platforms.
• Built a teacher dashboard from custom Superset charts, set the data refresh to every 15 minutes and tuned Superset for stability.
• Added a course-rating XBlock with its ratings dashboard, and sped up instructor analytics for a university-affiliated programme.

Problem

Accreditation points and every learner account lived in a rate-limited legacy LMS the client was leaving.

What I did

• Migrated 12,378 users and 237 courses through a resumable pull → transform → load pipeline (~50k API calls at 30 req / 10 s).
• Designed a separately deployed Django service talking to the platform over a Redis event bus.
• Delivered CME-point accrual reported to the accreditation body on certificate issue; mentored the client's in-house developer.

Problem

Learners jumped between the client portal, Open edX and a cyber-range with separate logins and no shared progress.

What I did

• Designed seamless SSO with the client portal as the OIDC provider and Open edX as the relying party.
• Built catalogue and progress sync for ~3,000 hands-on labs under strict API limits; connection pooling took p90 from 19 s to 0.56 s.
• Pushed enrolment and completion events back to the client's system of record.

Problem

The programme needed identity, roles and cohort management in front of an LMS it must not modify.

What I did

• Delivered learning groups with scheduled reconciliation against the LMS: drift detection, retries, convergence status.
• Built the authorization app: a permission catalogue with runtime-defined roles, targeting ≤100 ms p95.
• Authored 7 ADRs for identity, enrolment and messaging contracts.

Problem

Facilitator-led cohorts were needed on a platform whose core the client forbade patching.

What I did

• Led an 11-feature scope within a fixed hour cap: cohort API, invite codes, facilitator roles, cohort-scoped discussions.
• Found a grading defect that left 442 learners at 0% and 96 certificates unissued; shipped the fix and a self-healing completion rollup.
• Ran an AI-assisted delivery pipeline: spec → implementer agent → reviewer agent → local verification.