A risk check for under-construction homes in Uttar Pradesh. Search a UP-RERA project. TinyFish browser agents read six kinds of public records, and the app turns them into one NeevScore (300–900) with a source link for every point.

Live demo · Scope · Pages · How it works · NeevScore · Accounts · Credit protection · Configuration · Run it · Team

---

Live demo

Web app: https://neevcheck-up.vercel.app · API: https://neevcheck-api.onrender.com (health, docs).

Every check is live: nothing is mocked, and there is no demo mode. The API runs on Render's free plan and sleeps when idle, so the first search can take up to a minute while it wakes. To run it yourself, see Getting started; to deploy it, see docs/deployment.md.

The problem

Buying an under-construction flat is often a family's biggest cheque. The facts that decide whether it will be delivered are public, but they are scattered across government portals that few buyers know about, let alone read.

51,000	homes Supertech booked for delivery in 2010–12, many still waiting (Supreme Court, Feb 2026)
49,748	homes in 16 stalled Supertech projects handed to NBCC to finish (NCLAT order, Dec 2024)
UP-RERA	the one register NeevCheck reads in full: promoter, dates, extensions and complaint counts for every project

A broker's brochure says "possession on time". The records may say the completion date was extended three times. NeevCheck reads the records so the buyer doesn't have to.

Scope: UP-RERA only

NeevCheck covers Uttar Pradesh RERA projects, numbers like UPRERAPRJ8334. A number of another state's RERA, or a project in another state, gets one calm line: "NeevCheck covers Uttar Pradesh RERA projects for now. Check this project on its own state's RERA website." Nothing is searched for it.

Why not other states? UP-RERA is the one portal whose project pages a program can open and read: the details page, the project summary with its complaint counts, and the extensions granted. Other states' portals ask for a CAPTCHA or limit access, and NeevCheck never bypasses either. Thin, search-only coverage of them would look like a check and prove nothing, so it is left out.

Beyond UP-RERA, a check reads the company registry, group companies, insolvency orders, land authority lists, lender notices, news and court reports for the builder. A source it cannot read stays unread: that lowers confidence and never adds or removes points.

Pages

Page	What it is
Home /	The headline, the search box, More ways to find it, example searches, and your own recent checks
Check a project /check	Search, More ways to find it (builder, locality, brochure PDF, QR code, orders), saved reports of famous projects, what a check reads and what it costs
How it works /how-it-works	The steps, the six pillars and weights, the 300–900 scale and bands, what Not enough evidence yet means, and every source with its link
For RWAs /rwa	How a residents' group watches a project with Neev Sentinel, alerts with stop links, and a sign-in prompt
Alerts /alerts	The signed-in account's watches and alerts
Privacy /privacy	What is kept about a visitor, and how to remove it
Login /login	Optional Google sign-in
Report /report/<id>, My reports /reports, Compare /compare	A finished check, your saved checks, and projects side by side

Every page has an English and a हिंदी version, chosen with the EN / हिं switch. Old links to /#search, /#how-it-works and /#sentinel go to the new pages.

_{Login: optional. You never need an account to read a saved report.}

How it works

1. Find it. By name, UP-RERA number, builder, locality, a brochure PDF or a QR code. Many projects have no website and no number to hand; there is a path for them.
2. Confirm it. Candidates are listed with how sure we are of each number. Nothing runs until the buyer says "yes, this is the one".
3. War Room. Six lanes, read two at a time: UP-RERA, the company registry, group companies, insolvency and court cases, land authorities, news and lenders. Each lane streams its progress and says honestly when it stopped.
4. NeevScore. A deterministic formula turns the findings into a 300–900 score. Same records, same score.
5. Report. Score, reasons, evidence ledger, timeline and a buyer action plan, in English or हिंदी.
6. Sentinel. A weekly re-check alerts the buyer only when a record actually changes.

_{Report: the score, the reasons behind it and the proof for each.}

_{War Room: a recorded check of UPRERAPRJ8334, replayed. Six lanes, each with its own status.}

The screenshots and animations above come from a local production build that replays saved fixtures. They never call TinyFish.

Find a project

A buyer may know only a name, a builder's brand, a neighbourhood, a brochure or a hoarding's QR code. The search reads UP-RERA and the open web side by side, and never calls a listing site a source: listing sites only help find the number.

You have	What NeevCheck does
A name	Searches up-rera.in, and the open web with no domain limit. UPRERAPRJ numbers are found in the results, then voted on
A UP-RERA number	Opens its own up-rera.in page. If the page gives nothing, the same vote confirms the number exists
A builder	The builder's registered projects from up-rera.in and the open web, plus company-registry pages with CIN, each company kept apart
A city or locality	The place is added to the open-web search
A brochure or agreement	Upload a PDF (5 MB). It is read in memory, the UP-RERA numbers and the words around them come back with phone numbers, e-mail, PAN and Aadhaar removed, and the file is never stored
A QR code	Scan it with the camera or choose a photo. The browser decodes it; only the text inside is sent

The vote. A number printed by a page of up-rera.in earns 3 points, by the builder's own site 2, and by any other distinct domain 1. A number is kept at 3 points or when two domains print it, and only if one of them prints it next to the project's name. A number beside a different project's name does not count. Several numbers for one name (phases, towers) are separate candidates.

Status	Meaning
confirmed	A page on up-rera.in prints this number
listings_only	Builder and listing sites print it; no UP-RERA page was read. A check will try UP-RERA and say what it could not verify
from_document	Taken from the buyer's own PDF or QR code
from_order	Found in a public order that names the project, with a link to the order

When nothing is found. NeevCheck says only: "We couldn't find a RERA registration for this project in public records." It never calls a project illegal or unregistered, because a public search can miss a project. It then tells the buyer what to do: ask the builder for the number in writing, that a project of more than 500 square metres of land or more than eight apartments must be registered before it is advertised or sold (RERA Act, Section 3), not to pay until the number can be verified, and where to tell UP-RERA. A test bans accusing words in that card, in English and Hindi.

NeevScore (300–900)

NeevScore v2 is a number from 300 to 900: higher is safer. The backend owns the scale (GET /api/meta serves it; the frontend keeps no copy). It follows the shape of a credit report:

NeevScore = 900 − 6 × Σ (pillar weight × pillar penalty)        clipped to 300–900

Points come off only for records we find. Absence of a record never adds points. A source we cannot read becomes reason code R12 Data unavailable: it costs nothing and lowers confidence, the share of the six pillars that were read.

Pillar	Weight	Examples
Regulatory compliance (RERA)	30%	registration status, extensions, missed quarterly updates
Legal and insolvency	25%	insolvency of the promoter or this project, complaints, cases
Delivery track record	20%	complaints against the builder, other projects of the same group delayed or lapsed
Financial and corporate health	15%	struck-off or disqualified directors, charges, auction notices
Land authority and approvals	5%	dues lists, missing completion or occupancy certificates
Sentiment and news	5%	delay or protest coverage (secondary, capped)

Band	Range
Low risk	750–900
Watch	600–749
Elevated risk	450–599
High risk	300–449, and any hard stop (for example this project in insolvency per an official page) caps the score here
Not enough evidence yet	shown instead of a band, with a dash

Not enough evidence yet. A band is shown only when the UP-RERA record (the first pillar) and at least four of the six pillars were read. Below that there is no number, because a number nobody can stand behind is worse than none. A hard stop always shows.

Partly read pillars. The complaint counts on UP-RERA's own summary page speak for the builder (the delivery track record pillar) and for the project (the legal pillar). When they are all that was read for a pillar, the report says Partly read (UP-RERA complaints). They are a real record, so the pillar counts as read; the rest of it stays unread.

A real example. The recorded live check of UPRERAPRJ8334 (Supertech Eco Village-I Phase I-C, read on 25 Oct 2026), the report in the screenshot above, gave 495 · Elevated with 5 of 6 pillars read:

900  starting score
-150 15 complaints on UP-RERA about this project            (R05, primary)
-126 completion date 105 months after the promised one      (R03, primary)
-120 5 of 5 group projects past their completion date       (R13, primary)
  -9 three news stories on delays                           (R11, secondary)
 ───
 495  Elevated · confidence 83% (5 of 6 pillars) · no hard stop
      not read: land authority and e-Jagriti (R12), listed with a link to check by hand

The 2,803 complaints against the promoter across all its projects are also a finding (R13, "Builder complaint record", capped); here the group's delays already use up that pillar. Every line links to the page it was read from. The check is recorded in frontend/e2e/fixtures/report.json and replayed by the browser tests; the backend tests replay backend/tests/fixtures/check_8334.json. The weights are transparent expert priors, to be calibrated on labelled stalled-project lists. Full method: docs/scoring.md.

Sign-in, saved reports and alerts

Google sign-in is optional. Checking a project, reading a saved report and sharing one never need it.

• Separate accounts. Reports and alerts belong to one Google account. Another account on the same browser sees none of them, and nobody sees anyone else's checks.
• Saved reports. Signed in, every check you start or reuse is saved to your account. Signed out, they stay in this browser and can be moved into an account later.
• Alerts with stop links. Watch a project with Neev Sentinel and it is re-checked about every week, within the site's daily budget. An alert is created only when a record changed, in English or हिंदी, and appears on your Alerts page. When email is set up it is emailed too, and every email ends with a link that stops that one watch. An account can watch 3 projects.
• What is kept. A Google id, a name, an e-mail and the ids of your reports. Delete it all from the account menu. See /privacy.

Credit protection

Every live check and search spends TinyFish credit, so each has a limit. All of them are environment variables (the table below); days and months are counted in Indian time.

• Sign in to start a live check. A new live check needs a Google sign-in, and an account gets 2 a day. Opening, sharing and reusing saved reports stays open to everyone.
• Reuse of recent reports. A report finished in the last 7 days is returned to anyone instead of starting a new check, at no cost. Re-check now works once per project every 7 days, and the report shows Next re-check possible on … until then.
• Saved searches. A search answer (a name, a builder, a locality or a RERA number) is kept in the database for 7 days, so the same query never calls Search twice. Searches per address are limited to 15 a day signed out and 40 signed in, and the whole site runs at most 200 a day; a saved answer never counts, and when the site cap is reached the page says so calmly.
• Daily and monthly budgets. The whole site starts at most 5 live checks a day and 60 a month, Sentinel's re-checks included. When they are used up the site says "Today's live checks are used up. Saved reports are still open; new checks open again tomorrow." Searching and opening saved reports keep working.
• Sentinel is paced. There is no timer: at startup and on requests, at most once an hour, the API re-checks watched projects whose last check is older than SENTINEL_INTERVAL_HOURS, at most SENTINEL_CHECKS_PER_DAY (1) a day, and never when the site's daily or monthly budget is used up. TinyFish Monitors are off (SENTINEL_MONITORS=false), because they bill outside that budget.
• Invite-only switch. With INVITE_ONLY=true, only the e-mails in ALLOWED_EMAILS can start live checks; everyone else still reads saved reports.
• Lanes that cannot succeed are not run. Consumer cases (e-Jagriti) always show a CAPTCHA, so no agent is sent: the lane says Not checked automatically: the site needs a CAPTCHA and links to the site. The group lane runs only when UP-RERA gives the builder's exact company name. The insolvency lane reads its page with Fetch first and sends an agent only when the page speaks of an admission.
• A stop when credit runs out. If TinyFish reports no credit left or a rate limit, the rest of that check's lanes are skipped, with no retries, and the War Room says Paused: live-check credits are low.
• What a check costs. Each check stores how many Search, Fetch and Agent calls it made. GET /api/admin/usage (with ADMIN_TOKEN) returns today's and this month's totals and the average per check.

How NeevCheck uses TinyFish

Every source tries the cheapest way to read a page first and escalates only when it has to.

TinyFish	Used for
Search	Finding the project on the open web and on up-rera.in; company pages; lender notices; news; public orders
Fetch	Opening UP-RERA project pages, company-registry pages and PDFs, lender and land-authority documents
Agent	The last step, for a page that needs clicks, such as an insolvency order that speaks of an admission
Monitors	Optional and off by default (SENTINEL_MONITORS): watching a project's page and builder news; a change re-checks and alerts

When an agent meets a CAPTCHA, a login or a paywall it stops. The lane says Unable to verify, the report lowers confidence, and nothing is bypassed. Results are cached per source, entity and day, so a second project of the same builder reads its company records from the cache. The escalation ladder is in docs/architecture.md.

Architecture

Diagram source (Mermaid)

flowchart LR
    U([Homebuyer]) -->|find, confirm| FE[Next.js app<br/>Check · War Room · Report]
    FE -->|REST + Server-Sent Events| API[FastAPI backend]
    API --> GATE[Limits<br/>sign-in · budgets · saved searches]
    GATE --> RES[Resolver<br/>open-web vote · document · QR]
    GATE --> ORCH[Check orchestrator]
    RES --> TF
    ORCH -->|six lanes| TF[[TinyFish<br/>Search · Fetch · Agent]]
    TF --> R1[(UP-RERA)]
    TF --> R2[(Company registry)]
    TF --> R3[(IBBI · NCLT)]
    TF --> R4[(Land authorities)]
    TF --> R5[(Lender notices)]
    TF --> R6[(News · courts)]
    ORCH --> SC[NeevScore engine]
    SC --> DB[(SQLite / Postgres)]
    DB --> API
    CRON[Sentinel scheduler] -->|weekly re-check| ORCH

Loading

sequenceDiagram
    autonumber
    participant B as Browser
    participant A as FastAPI
    participant T as TinyFish
    B->>A: GET /api/projects/search?q=Mahagun Mywoods
    A->>T: Search (up-rera.in + open web), or a saved answer
    A-->>B: candidates with status and sources
    B->>A: POST /api/checks {rera_id, details_url}
    A-->>B: 200 a recent report, or 202 {check_id} for a new live check
    B->>A: GET /api/checks/{id}/events (SSE)
    A->>T: Search, Fetch, Agent per lane
    A-->>B: source.progress / finding.added
    A-->>B: check.completed {display_score}
    B->>A: GET /api/checks/{id}/report

Loading

More detail in docs/architecture.md.

Tech stack

Layer	Choice	Why
Browser agents	TinyFish Search, Fetch, Agent and Monitors	Reads real government portals like a person, with a live view we can show buyers
Frontend	Next.js 15, React 19, TypeScript (strict)	Server rendering for a fast first load, file-based routes
Styling	Plain CSS with design tokens, CSS animations	Easy to read; no runtime animation library
Fonts	Geist, Geist Mono, Noto Sans Devanagari (self-hosted)	Clean UI type with proper Hindi support, and builds that never fetch fonts
Backend	Python 3.12, FastAPI, Pydantic v2	Typed API, async streaming with Server-Sent Events
Documents	pypdf (PDF text), jsQR (QR codes, in the browser)	Small, pinned, offline; no OCR service
Accounts	Auth.js with Google, a short-lived signed token to the API	One login system; the API only checks a signature
Storage	SQLite (local), Postgres (production) via SQLAlchemy 2	Zero setup locally, one switch to deploy
Quality	pytest, Ruff, ESLint, tsc, Playwright with axe	Every change linted, typed, tested and checked for accessibility
CI and security	GitHub Actions, gitleaks, Dependabot	Tests on every push, secret scan over full history
Hosting	Vercel (web), Render (API)	Free tiers

Trust and ethics

• Public records only. NeevCheck reads official portals and published reports. No private data is collected; an uploaded brochure is read in memory and never stored.
• No CAPTCHA, login or paywall is ever bypassed. The source is marked Unable to verify, or Not checked automatically when we know it always shows a CAPTCHA.
• No source, no claim. Every finding links to where it came from and when it was read. A fact from news or another secondary source is badged From a secondary source, never Confirmed. Listing sites, orders and uploads only help find a project's number and builder; they are never scored findings.
• Absence adds nothing. A record we could not find never adds points; an unreadable source only lowers confidence.
• Never invented. There is no demo mode: every check is live, and tests replay recorded fixtures only.
• Decision support, not advice. NeevCheck is not a credit rating or legal advice. Verify important facts on the official portal.
• Your keys stay yours. API keys live in a local .env, never in the browser, logs or commits. See SECURITY.md.

Getting started

Every check is live, so you need a TinyFish API key. To put NeevCheck online, follow docs/deployment.md.

git clone https://github.com/Nitinmittal373/NeevCheck.git neevcheck
cd neevcheck
cp .env.example .env            # then set TINYFISH_API_KEY in .env

# backend
cd backend
python -m venv .venv && source .venv/bin/activate
pip install -r requirements-dev.txt
uvicorn app.main:app --reload    # http://localhost:8000/docs
ruff check . && pytest -q        # tests replay recorded responses, no key needed

# frontend (new terminal)
cd frontend
npm install
npm run dev                      # http://localhost:3000
npm run lint && npm run typecheck && npm run test:unit && npm run build
npm run e2e                      # browser tests; need `npx playwright install chromium` once

# after changing a backend model (from the repo root)
make contract                    # exports contracts/openapi.json and regenerates the frontend types

The key never leaves your machine: .env is git-ignored and the app never logs it. A TINYFISH_API_KEY set in your shell takes precedence over .env. For local work without Google sign-in, set REQUIRE_SIGNIN_FOR_CHECKS=false in .env.

Try it from the terminal once the backend is running:

curl "localhost:8000/api/projects/search?q=Mahagun%20Mywoods"
curl "localhost:8000/api/promoters/search?q=Mahagun%20India"
curl -X POST localhost:8000/api/extract -H "Content-Type: application/pdf" --data-binary @brochure.pdf
curl -X POST localhost:8000/api/checks -H "Content-Type: application/json" -d '{"rera_id": "UPRERAPRJ8334"}'
curl -N localhost:8000/api/checks/<id>/events               # live event stream
curl localhost:8000/api/checks/<id>/report
curl localhost:8000/api/limits                              # what a live check costs

The full API is described in docs/architecture.md.

To refresh the README pictures without spending any credit: build the app, start the recorded API (node e2e/mock-api.ts), point the site at it with NEXT_PUBLIC_API_URL, then run npm run capture in frontend/.

Configuration

Set the Render values on the API service and the Vercel values on the web app. Secrets are marked; their values are never shown here, never put in a NEXT_PUBLIC_ variable and never committed (.env is git-ignored). The full guide is docs/deployment.md.

Render (API)

Variable	Default	Secret	What it does
TINYFISH_API_KEY	required	yes	Your TinyFish key. Only the API ever sees it
FRONTEND_ORIGINS	http://localhost:3000		Web addresses allowed to call the API, comma separated
AUTH_BRIDGE_SECRET	unset	yes	Same value as on Vercel. Without it sign-in and saved reports are off
WEBHOOK_SECRET	unset	yes	Protects POST /api/sentinel/run and the monitor webhook
ADMIN_TOKEN	unset	yes	Sent as X-Admin-Token to read GET /api/admin/usage. Unset, that route is off
DATABASE_URL	SQLite file	yes	A Postgres URL keeps data across deploys
EMAIL_API_KEY	unset	yes	Key for the HTTPS e-mail API (Brevo or Resend) that sends alerts
SMTP_PASSWORD	unset	yes	Password for alert e-mail over SMTP, the fallback
REQUIRE_SIGNIN_FOR_CHECKS	true		A new live check needs a sign-in
CHECKS_PER_ACCOUNT_PER_DAY	2		Live checks one account may start per day
SITE_CHECKS_PER_DAY	5		Live checks the whole site may start per day
SITE_CHECKS_PER_MONTH	60		The same, per month
REUSE_REPORT_DAYS	7		A report this recent is shown instead of checking again
RECHECK_MIN_DAYS	7		Re-check now works once per project in this many days
SEARCH_CACHE_DAYS	7		How long a search answer is kept
SEARCHES_PER_IP_PER_DAY	15		Searches per address per day, signed out
SEARCHES_PER_IP_PER_DAY_SIGNED_IN	40		The same, signed in
SITE_SEARCHES_PER_DAY	200		Searches that really run across the whole site per day; a saved answer never counts
INVITE_ONLY	false		Only ALLOWED_EMAILS may start live checks
ALLOWED_EMAILS	empty		Comma-separated e-mails for the invite-only switch
FEATURED_RERA_IDS	empty		UP-RERA numbers for the gallery on the Check page; only those with a saved check show
RATE_LIMIT_PER_MINUTE	6		Check creation per address per minute
AGENT_CONCURRENCY	2		Browser agents at the same time
AGENT_TIMEOUT_SECONDS	300		Longest one agent may run
AGENT_SEND_MAX_STEPS	false		Send a step limit with each run (TinyFish beta accounts only)
FETCH_TIMEOUT_SECONDS	60		Longest one Fetch may take
SENTINEL_INTERVAL_HOURS	168		A watched project is re-checked once its last check is this old; 0 turns the schedule off
SENTINEL_CHECKS_PER_DAY	1		Re-checks the schedule may start per day
SENTINEL_MONITORS	false		Create TinyFish Monitors for watches. Off, existing ones are cancelled at startup
TRUST_FORWARDED_FOR	true on Render		Read the visitor's address from the proxy header
PUBLIC_BASE_URL	unset		The API's own address; only read when SENTINEL_MONITORS=true
SITE_URL	unset		The web app's address, for the stop link in alert e-mails
EMAIL_PROVIDER, EMAIL_FROM	brevo, unset		Which e-mail API to use and the verified sender. With EMAIL_API_KEY they turn e-mail alerts on
SMTP_HOST, SMTP_PORT, SMTP_USER, SMTP_FROM	unset, port 587		Fallback e-mail delivery of alerts over SMTP

Vercel (web app)

Variable	Default	Secret	What it does
NEXT_PUBLIC_API_URL	http://localhost:8000		The API's address, no trailing slash
NEXT_PUBLIC_SITE_URL	http://localhost:3000		The site's own address, for links and share previews
NEXTAUTH_URL	unset		The site's address, for Google sign-in
AUTH_SECRET	unset	yes	Long random string that signs the sign-in session
AUTH_GOOGLE_ID	unset		Google OAuth client id
AUTH_GOOGLE_SECRET	unset	yes	Google OAuth client secret
AUTH_BRIDGE_SECRET	unset	yes	Same value as on Render

Without the Google settings the site works as before and /login says sign-in is not set up.

Project structure

.
├── backend/                 FastAPI service
│   ├── app/
│   │   ├── api/             routes: search, promoters, extract, checks, events (SSE), reports, limits, admin, sentinel
│   │   ├── sources/         one module per public source, plus the shared agent runner
│   │   ├── states/          the resolver, the open-web vote, UP-RERA confirmation, builder and order search
│   │   ├── budget.py        the site, account and re-check limits
│   │   ├── search_gate.py   saved search answers and the per-address search limit
│   │   ├── usage.py         counts TinyFish calls per check and stops them when credit runs out
│   │   ├── extract.py       numbers in a PDF or QR text, with personal details removed
│   │   ├── scoring/         NeevScore engine (v2/) and the classic 0–100 score
│   │   ├── sentinel/        watches, change detection, alerts, email
│   │   ├── orchestrator.py  runs a check and streams its events
│   │   └── tinyfish.py      the only module that calls TinyFish
│   ├── scripts/             API contract export; recorders for test fixtures
│   └── tests/               pytest, with recorded TinyFish responses in fixtures/
├── frontend/                Next.js app
│   ├── e2e/                 Playwright tests against a recorded API
│   ├── scripts/             capture.ts: the README media, from recorded fixtures
│   └── src/
│       ├── app/             routes: home, check, how-it-works, rwa, alerts, privacy, login, report, reports, compare
│       ├── components/      pages/, find/ (More ways), confirm/, cam/ (War Room), report/ …
│       ├── lib/             API client and types generated from the contract, English and Hindi copy
│       └── styles/          tokens, globals, app screens and print
├── contracts/               openapi.json, exported from the backend (the API contract)
├── docs/                    architecture, scoring, data sources, deployment, assets
├── render.yaml              Render Blueprint for the API
└── .github/                 CI, issue forms, pull request template

Roadmap

• Phase 0 · Foundation — repo structure, CI, security policy, contribution guide, final design approved
• Phase 1 · Backend core — API, data models, NeevScore engine, live UP-RERA source, event stream, shared agent runner
• Phase 2 · TinyFish agents — source agents, live view streaming, CAPTCHA handling
• Phase 3 · Frontend foundation — design tokens, layout, animated skyline, home page
• Phase 4 · Check flow — search, confirm, Agent Cam, report
• Phase 5 · Sentinel and sharing — weekly re-checks, alerts, PDF, Hindi
• v2 · Credit-style score — NeevScore 300–900 with reason codes, War Room, family tree, evidence ledger, complaint draft, Monitor watch
• One contract, one score — the API contract is exported from the backend and generates the frontend types
• Uttar Pradesh only — one register read in full; other states answered with a calm line
• Find a project — open-web search with source voting, UP-RERA confirmation, builder and locality search, brochure and QR, public orders, and factual guidance when nothing is found
• Accounts — optional Google sign-in, separate saved reports and alerts with stop links
• Credit protection — sign-in for live checks, reuse of recent reports, saved searches, daily and monthly budgets, an invite-only switch, and a stop when credit runs out
• Real pages — Check, How it works and For RWAs have their own addresses, in English and हिंदी
• Phase 6 · Launch — deployment, calibration of the score weights, v1.0

Team

Built by Team RYZEN from KIET Group of Institutions for the HackIIITD 2026 finale at IIIT Delhi.

Member	Role	Focus
Nitin Mittal (team leader) github.com/Nitinmittal373	Backend	Backend services and integrations
Rishi Bhaskar github.com/rishibhaskar777	Full-stack	Frontend and backend
Siddharth Varshney	UI/UX	Interface and experience design
Shreyansh Gupta	Architecture and idea	System architecture, design and the original idea

Want to help? Read CONTRIBUTING.md.

License

MIT © 2026 Rishi Bhaskar and Nitin Mittal

_{Built on public records, read by TinyFish agents.}