QueueLess AI

QueueLess AI uses NVIDIA Nemotron on Nebius to end wasted civic errands with sourced document checklists, multi-stop planning, and zero-hallucinated queue wait times.

Comment

Inspiration

In emerging megacities like Lagos, Nigeria (home to over 15 million people), renewing a passport at the Nigeria Immigration Service (NIS) or getting a driver's licence from the Federal Road Safety Corps (FRSC) often turns into an ordeal lasting 8 plus hours and requiring up to 2 to 3 repeat trips. Citizens arrive at 6:00 AM only to find they brought the wrong state of origin certificate, paid at an unapproved bank, or arrived after the daily queue cutoff. When people turn to general-purpose LLMs (like standard ChatGPT), the models hallucinate: fabricating nonexistent appointment slots, reciting outdated fees from 2018, or guessing queue times out of thin air. We built QueueLess AI to solve this: a high-precision civic errand intelligence assistant that replaces hallucinations with strictly sourced official facts, intelligent multi-stop errand sequencing, and a mathematically enforced, zero-guesswork queue verification gate.

What it does

QueueLess AI helps citizens prepare for in-person government services so they can get errands done in fewer trips:

  1. Intelligent Intent Parsing: Citizens type natural-language requests (e.g., "I need to renew my 32-page passport in Ikeja and update my driver's licence"). The system extracts required agencies, document tiers, and geographical constraints.
  2. Grounded Sourced Checklists: Every document requirement, fee, and prerequisite links directly to verified government sources and SLA gazettes with check timestamps.
  3. Optimal Multi-Stop Sequencing: For citizens running multiple errands, the planner optimizes the sequence of visits based on opening hours and office locations.
  4. Zero-Hallucination Queue Gate: Queue wait times are displayed only when supported by verified, recent community observations. If data is stale or sparse, QueueLess explicitly states "Wait time unknown", refusing to synthesize false precision.
  5. Installable PWA & Open Radar Map: Works as an installable Progressive Web App on both mobile and desktop with offline document access, powered by keyless ESRI and OpenStreetMap tiles across 12 Lagos mainland and island districts.

How we built it

  1. NVIDIA Nemotron on Nebius AI Cloud: We integrated nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B via the Nebius Token Factory OpenAI-compatible API (https://api.tokenfactory.nebius.com/v1/). Nemotron handles intent extraction, parameter resolution, and structured JSON generation with rapid inference latency.
  2. Dual-Layer Validation: Model outputs are strictly validated server-side against JSON Schema 2020 (Ajv) before reaching the user interface.
  3. Modern Web Engine: Built on Next.js 15 (App Router) and React 19 with TypeScript, featuring a responsive 3-column Flow dashboard for desktop and an adaptive mobile dock.
  4. Open Geospatial Radar: Powered by Leaflet with free ESRI World Street Map and OpenStreetMap raster tiles completely eliminating the need for paid Google Cloud accounts or credit cards.
  5. PWA & Offline Shell: Fully offline-capable Progressive Web App with custom service workers and manifest shortcuts.

Challenges we ran into

  1. Eliminating LLM Civic Hallucinations: Standard prompts often yielded speculative fees. We designed strict Ajv JSON Schema 2020 guardrails to force the Nemotron model to select only validated service IDs from our verified corpus.
  2. Keyless Map Integration: Carto basemaps recently introduced watermarks requiring API keys. Because the pilot target had no Google Cloud billing access, we engineered a multi-tiered tile fallback using ESRI ArcGIS World Street Map and OSM France, achieving zero-cost, high-resolution mapping.
  3. React 19 / Leaflet Hydration: Leaflet's control container had timing conflicts during React 19 fast refresh. We built custom map invalidate-size hooks to ensure fluid rendering across viewports.

Accomplishments that we're proud of

  1. Seamless integration with NVIDIA Nemotron-3-Nano-30B running on Nebius Token Factory, achieving ultra-fast reasoning and JSON structuring.
  2. A grounded dataset of 12 verified government service centers across Lagos with verified coordinates and official requirements.
  3. A high-converting, ChronoTask-inspired modern landing page paired with an installable PWA for any phone or PC. 4.100% test pass rate across order planning, schema validation, and queue gate logic.

What we learned

Civic AI must prioritize truth over confidence. Stating "Wait time unknown" builds vastly more trust than presenting a confident, hallucinated 45-minute wait that leaves a citizen stranded at a closed gate. Using NVIDIA's reasoning models with strict JSON schema constraints on Nebius cloud provides the ideal balance between language flexibility and structural integrity.

What's next for QueueLess AI

  1. Crowdsourced SMS / USSD Reporting: Enabling non-smartphone users across Nigeria to submit and receive wait updates via SMS.
  2. Expansion to Abuja & Port Harcourt: Extending the verified government office directory to other major Nigerian hubs.
  3. Fine-Tuning Nemotron on Civic Gazettes: Using Nebius AI Cloud to fine-tune Nemotron on West African regulatory documents and state circulars.

Built With

Share this project:

Updates

Submission history