Confidential Proposal · Benchmarked
Real-Time Risk Event Map
Prepared for RMI. This document is access-restricted. Enter the access key to continue.
Incorrect access key. Please try again.
Benchmarked Group d.o.o.Ljubljana · EU
Prepared for RMI. This document is access-restricted. Enter the access key to continue.
×
BENCHMARKED
A live, automated intelligence layer that continuously ingests open sources, enriches every signal with AI, and plots verified risk events on a single operational map - built for RMI by Benchmarked.
Risk-relevant information is scattered across social platforms, messaging channels, government bulletins and news feeds - too fast and too noisy for manual monitoring. This platform automates the entire loop: it collects from five source families, uses large language models to classify and structure each item, maintains a live database of events, and presents them in real time on an interactive map and a prioritized event list. The result is a single pane of glass for operational situational awareness, with full transparency into which sources are currently online.
Interactive map with geo-located events, a live prioritized feed, and a connector health bar showing the real-time status of every source - so operators always know exactly how complete the picture is.
Illustrative mockup · sample data shown · final layout, regions and event taxonomy tailored to RMI requirements.
Every connector feeds the same live surface. This view shows the five open-source connectors streaming signals down onto the map, where each event is plotted as a severity-coded marker - red for critical, amber for elevated, grey for low or resolved.
Conceptual 3D view - connectors (top) continuously push enriched events onto the live risk surface (bottom).
Three integrated layers - collection, AI enrichment, and visualization - delivered as one cohesive platform.
Five purpose-built connectors continuously pull risk-relevant content from across the open-source landscape, on a scheduled polling cycle.
Every collected item is passed through LLMs that classify the incident, extract a clean title, location and severity, and de-duplicate against the live database.
Structured events feed a live interactive map (where coordinates exist) and a prioritized, filterable event list - with real-time connector status.
Each connector is independently monitored, rate-limit aware, and individually maintainable.
For every detected event, the LLM pipeline extracts and maintains a structured record:
| Field | Description |
|---|---|
| incident_type | Classified category (e.g. roadblock, security incident, civil unrest, infrastructure, weather). |
| title | A clean, human-readable summary of the event generated from the raw source content. |
| reported_at | Timestamp of when the event was first detected / reported. |
| duration | Whether the event is ongoing, and an estimate of how long it has been / will be active where determinable. |
| location | Geographic location and coordinates - when present in the source (see coverage notes below). |
| severity | AI-assigned severity level used for prioritization and map colour-coding. |
A continuous, automated pipeline running on a fixed cycle - from raw source to live map.
5 connectors poll their sources on schedule and capture new content.
LLMs classify, summarize, geolocate and score each item.
Events are de-duplicated and written to a live database.
Map + prioritized feed update in real time, with connector health.
A platform that depends on third-party sources is fundamentally different from a static website. It must be actively maintained to keep working. We want to be completely transparent about this before you commit.
Source platforms regularly change their APIs, page structures, authentication, and access rules - often without notice. When they do, the affected connector breaks until it is updated. We budget approximately 5 hours per month to keep connectors healthy and adapt to these changes. Without this upkeep, connectors will gradually degrade and go offline.
No automated collection system captures everything. There will always be events the platform misses or surfaces late. This is inherent to the problem, not a defect, and should be planned for accordingly.
Synchronization windows mean events appear only on the next polling cycle; API rate limits cap how much we can pull in a given period; platform anti-scraping protections can block or throttle collection; and content that is private, removed, geographically untagged or ambiguous cannot be reliably captured or placed on the map.
Rather than hide these limitations, the console shows the live status of every connector - online, degraded, or offline - so operators always know how complete the current picture is and can interpret the map accordingly.
The platform is designed to dramatically improve situational awareness and reduce manual monitoring - but it complements, and does not replace, human judgement and other intelligence sources.
Optional additions that can be scoped as future phases once the core platform is live.
A fixed price for the core build, plus a recommended maintenance retainer to keep the system reliable over time.
≈ 5 hours / month to maintain connectors, adapt to platform changes, and keep the system healthy. Strongly recommended - see considerations above.
LLM API usage, hosting and collection infrastructure - billed at cost and dependent on volume. Estimated range provided for planning.
All amounts are net (excluding VAT). The maintenance retainer and operating costs are separate from the one-time build. Scope extensions are quoted individually. This proposal is valid for 30 days from the date of issue.
Confirm regions, source accounts/channels, incident taxonomy and severity definitions with RMI.
Build and test the five connectors and the LLM classification / geolocation pipeline against live data.
Build the live map, prioritized event list and connector health dashboard.
End-to-end validation, tuning, deployment and walkthrough with the RMI team.
We're happy to walk the RMI team through this proposal, refine the scope, and align on regions and sources before kick-off.