18-year-old full-stack developer building reliable systems · Codeforces Expert · 451 problems solved
LinkedIn · Unstop · Email · LeetCode · Codeforces
I am an 18-year-old developer and CTO at Vedonyx, based in New Delhi, India. I work primarily with TypeScript, Python, React, FastAPI, PostgreSQL, and Flutter. I enjoy taking products beyond the demo stage: defining system boundaries, testing failure cases, documenting trade-offs, and deploying software that other people can evaluate.
Unstop: Top 3% in Unstop Rankings (November 2026).
| Project | Engineering highlights | Stack |
|---|---|---|
| Backend Fieldguide · Live course | Interactive ecommerce backend course with 108 lessons, 16 practice missions, and browser simulations for inventory and payment retries. | Next.js, React |
| RollbackReady · API | Evidence-based Prisma migration review with deterministic checks, disposable PostgreSQL simulations, failure timelines, verified recovery plans, CI, containers, and cloud deployment. | Next.js, TypeScript, FastAPI, PostgreSQL, Docker, Kubernetes |
| Prisma Schema Visualizer · Live | Interactive ER diagrams with relationship routing, obstacle avoidance, focus states, zoom, pan, and PNG export. | Next.js, React, TypeScript |
| Memvora CLI | Installable CLI for terminal capture, content hashing, offline queueing, workspace configuration, and authenticated synchronization. | Python, Typer, HTTPX |
| VectorBridge | Local-first CDR/SVG conversion interface with engine discovery, temporary-file cleanup, progress feedback, and light/dark themes. | TypeScript, Next.js, Node.js |
| Tool | What it solves |
|---|---|
| envproof | Finds missing, empty, duplicate, and undocumented environment keys without exposing values. |
| jsonshape | Detects breaking structural changes between JSON API payloads. |
| portpick | Finds or safely reserves predictable TCP ports for local tools and test suites. |
| PR Size Guard | Adds configurable pull-request size feedback without tokens or noisy bot comments. |
| Stack Trace Sprint | Turns realistic failure traces into a keyboard-accessible debugging game. |
| Repo Radar | Audits repository health and produces actionable local or CI reports. |
| API Pulse | Checks endpoint availability, accepted statuses, and latency targets from a small contract. |
| Contrast Lab | Tests accessible colour pairs with exact WCAG calculations and shareable URLs. |
Browse the complete project index →
- Hindsight: count shared source tokens once per recall — reported and benchmarked #6329 and supplied the original patch. The merged fix reuses token counts across observations and skips counting when the budget is unlimited. The maintainer measured 8.42× faster source selection (6.40 ms → 0.76 ms across 100 observations sharing 100 sources, with a 256-token budget per observation), preserving the selected facts, their order, and the truncation flag.
- Hindsight: narrow scoped entity reads before counting mentions — reported and benchmarked #6330, credited by the maintainer in the merged PR. The shared visibility query now reads the requested entities' mentions when below a cap, retaining the broad-query fallback. The maintainer's benchmark measured about 10× faster narrow name searches (585 ms → 58 ms; 300,000 memories, 20,000 entities), with the same rows and counts.
- Hindsight: share freshness checks between pages with the same filter — supplied the report, benchmark and patch idea in #6325. The merged fix queries each shared filter once while preserving each page's cutoff; the maintainer's benchmark measured 51.64× faster shared-filter freshness checks (903.7 ms → 17.5 ms; 53 pages, 100,000 memories, PostgreSQL 18 with JIT off).
- Hindsight: look up recalled observations by ID — contributed the patch in #6324, merged with co-author credit. Reused a dictionary for delete, merge-target and update lookups; the original local benchmark measured up to 6.69× faster Python bookkeeping (266.21 µs → 39.77 µs; 240 observations, eight deletes and eight updates), excluding database and model calls.
- Hindsight: probe for pending consolidation without counting the bank — contributed the original patch in #6326, merged with co-author credit. Replaced full-bank counts with an indexed pending-work check; the maintainer's warm-cache benchmark measured about 11.7× faster checks (≈140 ms → ≈12 ms; one million rows, 100,000 failed facts), excluding consolidation itself.
- Hindsight: count shared tag lists once when listing observation scopes — supplied the original patch in #6328, merged with co-author credit. Grouped repeated tag lists before sorting and reused the result for the page and total; the maintainer's benchmark measured up to 29.53× faster scope listing (3.16 s → 107 ms; one million observations, 53 tag sets), with identical output.
- Hindsight: batch document-tag reads during scoped retain — supplied the report, benchmark and original patch in #6327. Merged upstream with maintainer credit; the original local benchmark measured 35.89× faster tag reads and permission checks for 3,500 documents (3.04 s → 84.82 ms), excluding extraction and writes.
- Hindsight: avoid repeated scans when deleting knowledge folders — reported and supplied the original patch in #6323, acknowledged by the maintainer in the merged PR. The merged fix keeps rows on the traversal stack. The original local Python-walk benchmark measured up to 393.39× faster (10.710 s → 27.22 ms; 25,000 bank nodes, 2,000 folder nodes), excluding database operations.
- Wonderland: Browser element picker for Component Preview (#74), Commit/Sync main action (#68), controls in narrow and short panes (#66), file browsing and renaming (#59), closed-tab recovery and relative paths (#53), log copying and shared-view synchronization (#61), recent files and keyboard search (#56), workspace bounds and tab dragging (#51), startup settings preservation (#50), safe replace undo and search positions (#48).
- Babel: speed up MO file writing — used bytearray buffers to avoid repeated string-table copying, with byte-for-byte output and error compatibility checks.
Show 11 more merges (11–21)
- tomlkit: speed up array appends — avoided rebuilding the full index on normal array appends, preserving formatting behavior and compatibility fallbacks, with regression coverage.
- fsspec: speed up TAR file listing — avoided repeated directory-cache scans, with up to 221.795× faster warm TAR listings in Django-source benchmarks (476.572 ms → 2.149 ms), plus compatibility fallbacks and regression tests.
- Babel: speed up PO location lookups — optimized duplicate checks for PO source references while preserving reference order, with regression coverage.
- OpenClaw: avoid Doctor backups for unchanged one-shot jobs — made cron repair a no-op for canonical schedules while preserving backups for real changes.
- OpenClaw: preserve request context in queued operations — kept each queued operation's diagnostic context tied to its submitting request, with regression coverage.
- OpenClaw: keep distinct mixed-script memories in search results — improved Unicode tokenization so distinct Cyrillic, Arabic, and Indic notes are not treated as duplicates.
- OpenClaw: find tool parameters inside composed schemas — fixed tool discovery for parameter names and descriptions nested in
anyOf,oneOf, andallOf, with regression tests. - OpenClaw: seed Discord presence data after startup and reconnect — merged with focused regression coverage and runtime-boundary validation.
- OpenClaw: clarify replies to buttonless question prompts — replaced tap-only guidance with instructions that also work in text-only channels.
- OpenClaw: correct node-approval command examples — merged documentation fix.
- Kubernetes: update the gRPC dependency — merged upstream contribution.
- OpenClaw: localize the Security settings heading — proposed the original fix in PR #153718. Maintainers merged replacement PR #191143 with co-author credit, making the heading follow language changes while preserving help-text fallbacks.
- OpenClaw: handle Windows paths in repository checks — proposed the original fix in PR #168956. Maintainers independently implemented and merged replacement PR #190069 with co-author credit, correcting Windows path handling in SQLite and changelog checks.
- XlsxWriter: speed up object positioning on uniform-height sheets — reported and benchmarked issue #1210, then submitted PR #1211. The maintainer landed a simpler implementation directly on
mainwith co-author credit; the PR itself was closed without merging.
- Product engineering: accessible interfaces, typed APIs, authentication, persistence, and responsive interaction.
- Backend systems: Python services, PostgreSQL, migrations, validation, authorization, and failure-safe workflows.
- Delivery: automated tests, GitHub Actions, containers, Vercel, Cloud Run, and Kubernetes.
- Problem solving · 451 problems solved, including 392 on LeetCode and 59 on Codeforces. Codeforces Expert with a rating of 1777. LeetCode · Codeforces
| Area | Tools |
|---|---|
| Languages | TypeScript, Python, JavaScript, Dart, SQL |
| Product | Next.js, React, Node.js, Flutter, FastAPI |
| Data | PostgreSQL, Prisma, SQLAlchemy, Pandas, NumPy |
| Delivery | GitHub Actions, Docker, Kubernetes, Vercel, Google Cloud |
Open to collaborating on developer tools, data products, and reliability-focused web systems.




