Team

Team X

Project Concept

No description has been added yet.

Entry

Status: Submitted

Last saved: May 09 at 5:19 PM PDT

View Entry

Team Roster

Message board not available for this team yet.

Devanshi Prajapati Team Lead RSVP Approved

Software Engineer at Elas

Owned the Next.js 16 / React 19 / Tailwind v4 frontend and the agent wiring. Replaced CopilotKit's <CopilotChat> with a custom generative workflow surface (CommandBar, Stepper, GenerativeForm, LineItemsEditor, ApproveSendButton) so users can edit forms inline instead of chatting. Stood up the CopilotKit v2 runtime (@copilotkit/runtime/v2 + BuiltInAgent) against Google DeepMind's Gemini 2.5 Flash via Google AI Studio, along with the CopilotKit A2UI middleware and the basic catalog. Drove the agent programmatically using useAgent + copilotkit.runAgent + subscribeToAgentWithOptions, so the command bar can fire a single prompt and stream tool calls without a chat thread. Registered six frontend tools (advance_step, lookup_record, update_estimate, approve_send, onboard_company, save_submission) via useFrontendTool with Zod schemas and pushed company/workflow/flow/system-prompt context to Gemini via useAgentContext. Authored the typed REST client (lib/api.ts) mirroring Devanshi's Pydantic models and the CompanyContext hydration that loads everything on mount.

Hey! I'm a recent new grad who has been completely obsessed with AI since it came around.

Looking to learn more about implementing AI in existing systems and building fully agentic platforms for clients.

1. Creating custom OCR models for conferences that can assist with contact capturing. 2. Developing a knowledge base architecture using graphs that can assist with provided better context for smater answers

Dhairya Gundechia RSVP Approved

Data Scientist at Elas

Owned the FastAPI + Pydantic v2 backend and the Supabase schema. Wrote the three-step migration set (001_init, 002_company_workflows, 003_multi_workflow) that enables a single tenant to configure multiple workflows side by side. Built every REST route the frontend depends on: /company/{id}/status, /company/{id}/onboard, full /workflow/... CRUD, /flow/... GET/POST/PATCH (the merge-patch path was critical so the agent could update one prop without wiping the rest), /submissions/... with step + customer filters, plus the two action endpoints /actions/estimate-update (Decimal-clean total recompute) and /actions/approve-send (idempotency-keyed reserve in Supabase). Also authored the Workflow.to_system_prompt() projection that turns a workflow definition into Gemini-readable context, and the pytest + httpx test suite covering each route.

Dhairya Gundechia is an AI Engineer at elas, focusing on cloud-native AI products and full-stack ML systems using FastAPI, TensorFlow, and AWS. Their current projects include CulinaryVertex, a real-time AI application integrating Gemini 2.0 Flash and LiveKit for multimodal interactions, and MatchPoint and HousingInMexicoBrazil, built with Next.js and TypeScript. Dhairya has 3 years of experience, with past technical implementations featuring email spam classifiers, MNIST CNNs, and stock price prediction APIs. They studied Software Engineering Systems and Information and Communication Technology at Northeastern University and Dhirubhai Ambani University. Dhairya is open to intros and prefers direct messages.

Technical architecture, cloud-native AI products, full-stack ML systems, real-time multimodal applications, LLM fine-tuning, production-grade agents, FastAPI, TensorFlow, AWS, React/Next.js, TypeScript, LiveKit integration.