Vaibhav Lalwani / SYSTEMS·26 Résumé
∞ / Archive

Everything shipped.

37 builds, newest first: production systems, hackathon entries and open source. Eight have a full case study; the rest link to the strongest available demo, release or repository.

JUDGES’ PICKYUZU01

Agent commerce · SharedOS · Judges’ Pick

Yuzu

The market where agents hire agents. A goal and budget become discovery, competing bids, capability proof, negotiation, a scoped contract, verified delivery and settlement with signed receipts.

LOCAL CREATOR OSVYREALM02

Creator tools · Local-first

VYREALM

A Windows creator studio for turning footage and ideas into editable projects with timeline editing, captions, narration, local storage, real video export and optional MCP/model workflows.

EXACTLY ONCEREQKEEPER03

Payments · Agent safety

ReqKeeper

Exactly-once settlement for Request Network obligations through KeeperHub. Concurrent workers can retry the plan without retrying the money.

VOICE-FIRST STUDYVIVA04

Voice AI · AssemblyAI Hackathon

VIVA

Study out loud. Dictation and live streaming turn spoken uncertainty into source-grounded feedback and an active-recall queue for what the learner actually missed.

05

Agents · Buildathon · Winner

Leverage

An intelligence resource manager. Your best model keeps the plan, an auction hires cheaper models per task, a checkpoint survives every worker failure, and the repository's own tests decide what counts. RocketRide × SCU Buildathon winner.

06

Agents · WebMCP

Cherry

The user-owned apprenticeship, memory, mission and verification layer for AI agents: skills a person can inspect, receipts a person can verify. WebMCP Challenge 2026 entry.

07

Documents · Buildathon

Atkin

The document decision desk. Screening, evidence extraction with quote verification and clean exports for anyone facing a stack of documents and a deadline. RocketRide × SCU Buildathon entry.

08

Evals · Release gate

XACT

Prove the fix. Turns one bad AI response into a regression test that never fails twice: a complaint-to-regression release gate for AI agents, built on RocketRide.

09

Agent memory · HydraDB

Lacuna

Temporal, provenance-first memory for AI agents on HydraDB. What is true now, what changed, which sources disagree and where the evidence is, over web, CLI, API and seven read-only MCP tools. Hack Hydra 2026, Track 3.

10

Retrieval · Winner

QueueProof

Evidence-backed cross-source retrieval. Ask one question across your tools and every claim in the answer opens back to the record it came from. HydraDB × Connectors Hackathon winner.

11

Developer tools · Agents

skillister

Find, route and install agent skills across any agent CLI. One index of about 1,200 skills serving five hosts, with an audit that says what a skill actually does.

12

Fintech · Onchain

FinalTab

Receipt to verified on-chain settlement. Groq vision reads the receipt, a deterministic money engine splits it, and an EIP-3009 batch settlement is executed and chain-verified on Base Sepolia.

13

i18n · Verification

BhashaFix

Open-source localisation release engine: real browser evidence, bounded repair and identical-test verification across web, CLI, MCP and CI.

14

Data · SRE agent

Firebreak

A data-incident SRE agent on DataHub. Finds the blast radius of a bad dataset, drafts the fix and keeps the receipts for the review.

15

Integrity · Agents

Veritas

An integrity agent for agent outputs: SHA-256 digests and Ed25519 attestations anyone can verify. NANDA hackathon entry.

16

Agent security · Winner

HydraSentry

A memory-integrity firewall for AI agents. Catches a poisoned memory before the agent acts and seals a signed certificate. HydraDB Build Blitz winner.

17

Simulation · MSc

Kodro

An offline robot design and simulation studio. Build a robot, program it, and watch it work in a live 3D world, all local with no cloud.

18

Local-first AI

Own Wiki

A desktop second brain powered by a local LLM. RAG chat over your own files, a 3D knowledge galaxy, and a multi-agent research swarm, zero cloud.

19

Multi-agent · Commerce

DRIP

An AI sneaker stylist. Multi-agent styling that reasons about fit and taste, with live StockX resale prices and buy links per shoe.

20

Memory agent

Continuity

A self-evolving memory agent on HydraDB. Remembers across sessions, supersedes itself on contradiction, and warns you when recalled context changes the safe answer.

21

Agents · Fintech

Recoup

An AI money-recovery agent. Scans real Gmail receipts, grounds every claim in consumer-protection law, and stops at a human approval gate before it acts.

22

Code review · Graph

Keystone

Governed change review on the GitLab Orbit code knowledge graph: deterministic blast radius, precedent at the moment of decision and a tamper-evident audit ledger. GitLab Transcend Hackathon entry.

23

Agents · Trust and safety

PitchCraft

An agent that flags ticket-resale scams. Gemini and Google ADK grounded in MongoDB Atlas Vector Search and the MongoDB MCP server. Google Cloud Rapid Agent Hackathon, MongoDB track.

24

Agents · SRE

RecallOps Cortex

A recall-containment command center. A live openFDA recall scoped across BigQuery into a blast radius, with drafted actions behind an audit chain.

25

Multi-agent · Civic

CIVICTAS

A multi-agent civic decision copilot. Gemini and Groq providers with a mock mode so the demo runs offline. USAII Global AI Hackathon 2026 entry.

26

Agents · Platform

Praxon

Open-source AI agent platform: a multi-LLM router over free providers, MCP-native, local-first, deployable on Vercel and Supabase.

27

Sales · Chatbot

Satonic

Open-source AI lead-generation chatbot, a free alternative to Apollo.

28

Memory · Productivity

RecallOS

An AI memory and execution system for the things you keep saving and never use. Stop saving, start building.

29

Content · MCP

ReleaseRail

An AI content engine with an MCP surface: 27 tools, resume tokens and 138 tests. AI Content Engine Hacks entry.

30

Dashboard · LLM ops

GLASSBOX

A live dashboard of AI conversations, built as a working product for a full-stack exercise. Simulated data by default, or bring your own Groq or OpenRouter key.

31

Retrieval · Finalist

Qyntra

A personal knowledge OS. Ingest your files, explore them as a 3D galaxy, and ask questions answered from your own material. Wikithon '26 finalist.

32

Multi-agent · OS

DelOS

A browser-OS where agents flow under pressure. An xterm agent fleet on isolated HydraDB tenants, coordinated so agents don't collide under load.

33

AI · EdTech

Cogniloop

A Socratic study system. Grades free-form answers against a rubric and tracks concept mastery, sub-second on Groq Llama 3.3 70B.

34

Medical AI · RAG

MedReviewAI

A grounded medical paper analyzer. Searches 13 academic sources, extracts PICO fields, and keeps generated claims tied to verbatim source evidence.

35

Research · Extension

ScholarAI

A Chrome extension that turns research papers into interactive AI sessions with summaries, extracted claims, and Q&A over the paper.

36

MCP · Registry

MCP Marketplace

A searchable registry of Model Context Protocol servers with one-line install snippets for Claude Desktop, Cursor and Claude Code, synced daily.

37

Web · Just for fun

The Film Theory Vault

Search any movie and the page regenerates in a cinematic style, auto-themed to the film's colours, with AI-written fan theories behind a spoiler gate.

Back to selected work