The AI Market Intelligence Copilot is a conversational AI system powered by Claude Sonnet 4.5, accessed via the Databricks Foundation Model API (FMAPI). It has access to 51 domain-specific tools spanning market data, forecasting, trading, risk, settlement, DNSP operations, and ML inference.
The copilot is the primary interface for analysts and operators who want answers to complex questions that require synthesising data from multiple sources.
User Message (React frontend)
Agentic Loop (up to 5 rounds)
├─── Claude Sonnet 4.5 (Databricks FMAPI)
├──────────────────────────────────┐
│ tool_call event (SSE) │
get_latest_prices get_stpis_metrics explain_price_event
Gold Table Gold Table Gold Table + LLM
(Lakebase) (SQL Warehouse) analysis
▼ tool_result event (SSE)
Claude synthesises results
Streaming response to user
Tool Description get_latest_pricesCurrent spot prices for all or specified regions get_price_historyHistorical prices with region/date/interval filters get_generation_mixCurrent or historical generation by fuel type get_interconnector_flowsCurrent interconnector flows and utilisation get_active_constraintsCurrently active AEMO constraint sets get_fcas_pricesCurrent and historical FCAS prices for all 8 services get_weather_dataTemperature, humidity, wind speed for NEM regions get_market_briefToday’s or historical daily NEM market brief
Tool Description get_price_forecastML price forecast (30-min, 4h, 24h horizons) get_demand_forecastML demand forecast for specified region and horizon get_weather_forecastBOM NWP forecast (next 72 hours) get_wind_forecastWind generation forecast get_solar_forecastSolar generation forecast (utility + rooftop)
Tool Description explain_price_eventIdentifies likely drivers of a price event (spike, trough, separation) compare_regionsSide-by-side comparison of specified regions on a metric get_market_summaryHigh-level market summary for specified period search_market_rulesSemantic search over AEMO rules and procedures (Vector Search)
Tool Description get_trading_signalsCurrent algorithmic trading signals get_forward_curvesForward price curves from ASX futures get_deal_positionsCurrent portfolio positions get_mtm_valuationCurrent MtM for portfolio or deal get_bid_stackCurrent merit order and bid stack get_battery_dispatchBESS dispatch status and revenue
Tool Description get_varValue at Risk (historical simulation or Monte Carlo) get_greeksOptions Greeks for positions get_credit_exposureCounterparty credit exposure get_position_limitsPosition limit utilisation run_scenarioCustom price scenario and portfolio impact
Tool Description get_portfolio_summaryPortfolio P&L, MtM, and key metrics get_pnl_attributionP&L attribution by driver get_realised_pnlRealised P&L for period
Tool Description get_settlement_summarySettlement balance and true-up by run get_prudential_statusCurrent prudential deposit and requirement get_compliance_calendarUpcoming regulatory deadlines
Tool Description get_stpis_metricsSAIDI/SAIFI/MAIFI and revenue impact get_asset_health_scoresAsset health index and failure predictions get_vegetation_risk_scoresML vegetation risk by span or zone get_hosting_capacityDER hosting capacity by network zone get_workforce_forecastML workforce demand forecast
Tool Description get_aio_compliance_statusAIO obligation tracking calculate_rab_rollforwardRAB projection with configurable assumptions get_rab_sensitivityWACC and inflation sensitivity analysis get_dnsp_benchmarksAER benchmark comparison get_contractor_scorecardContractor performance metrics generate_aio_draftClaude drafts AIO submission section get_dapr_inputsDAPR data compilation get_outage_schedulePlanned network outage calendar get_hosting_capacity_detailDetailed hosting capacity by zone get_der_pipelineDER connection application pipeline get_skills_gapWorkforce skills gap analysis
Tool Description get_generator_infoRegistered generator details from AEMO
Total: 51 tools
The copilot implements a multi-round agentic loop:
# Simplified agentic loop
async def run_copilot ( messages : list , tools : list ) -> AsyncGenerator:
while round_count < max_rounds:
response = await claude.messages. create (
model = " databricks-claude-sonnet-4-5 " ,
async for event in response:
yield SSEEvent ( type = " text " , data = event.text )
elif event.type == " tool_use " :
yield SSEEvent ( type = " tool_call " , data = event.tool_name )
result = await execute_tool ( event.tool_name , event.tool_input )
yield SSEEvent ( type = " tool_result " , data = result )
messages. append ( { " role " : " tool " , " content " : result} )
if response.stop_reason == " end_turn " :
break # No more tool calls needed
yield SSEEvent ( type = " done " )
User: Why did SA1 spot price spike to $12,000/MWh this afternoon?
Copilot: [calls explain_price_event, get_generation_mix, get_active_constraints, get_weather_data]
Claude: The SA1 price spike to $12,000/MWh between 16:30 and 17:00 AEST today
was driven by a convergence of three factors:
1. **Wind output collapse**: SA1 wind generation fell from 1,240 MW to 380 MW
in the 60 minutes prior to the spike — a 69% reduction triggered by a wind
change moving through the Adelaide plains.
2. **Binding interconnector**: The V-SA interconnector was at 100% utilisation
(importing 630 MW from Victoria). The N^^NIL_NVVT_29 constraint was binding,
preventing additional import to offset the wind loss.
3. **Hot afternoon demand**: Adelaide reached 38°C, pushing SA1 demand to
2,150 MW — 12% above the seasonal average for this time of year.
With wind down, interconnectors at limit, and demand elevated, only the most
expensive OCGT peakers remained available — Quarantine Power Station bid its
remaining capacity into bands 6 and 7 (>$10,000/MWh), setting the dispatch price.
User: Which Ausgrid assets are at highest failure risk this month?
Copilot: [calls get_asset_health_scores with dnsp=ausgrid, min_probability=0.3]
Claude: The ML failure prediction model has flagged 23 Ausgrid assets with
>30% failure probability in the next 12 months. The top 5 by risk are:
1. **TXF-NW-0847** (Zone substation transformer, Blacktown): 68% failure probability.
Built in 1974, health index 22/100. Last oil test showed elevated furans
indicating advanced insulation degradation.
2. **TXF-SW-1204** (Distribution transformer, Campbelltown): 61% failure probability...
The copilot endpoint (POST /api/chat) streams Server-Sent Events:
Event Type Payload Description text{"delta": "..."}Incremental text from Claude tool_call{"tool": "...", "input": {...}}Tool being called tool_result{"tool": "...", "result": {...}}Tool execution result done{}Stream complete error{"message": "..."}Error occurred
