Candlery extracts trading signals from corporate filings.
Deterministic translation layers bridging unstructured events and zero-hallucination trade execution.
The Execution Pipeline
How we process unstructured reality into definitive execution signals.
Raw Ingestion
Sub-millisecond scraping
NLP Parsing
Semantic extraction
Risk Firewall
Rule-based filtering
Execution Compiler
Automated routing
1. Raw Ingestion
Distributed AWS Lambda clusters poll SEC EDGAR and Indian BSE/NSE feeds. We intercept announcements within 50ms of the raw document hitting the Exchange servers, discarding non-material disclosures instantly.
The Risk Firewall
Preventing hallucination risk using hard-coded parameters, drawdown checks, and historical analogs.
Firewall Gatekeeper
Test how signals get filtered
Select a scenario payload to run through the firewall validator:
Evaluates prospective trade size against active portfolio drawdown limit (<2.5%).
exposure_limit <= portfolio.max_drawdownRejects any NLP parsed intent with confidence metrics below the hard boundary.
signal.nlp_confidence >= 0.95Validates target asset bid-ask spreads to prevent slip damage during market orders.
asset.avg_spread <= MAX_SLIPPAGE_TOLERANCEVerifies the extracted signal against historical corporate announcement outcomes.
db.find_historical_analog(signal.type, signal.intent)Hhalts routing if target asset trading volatility is in top 1% standard deviations.
asset.volatility < asset.volatility_rolling_avg * 2.5Cancels trade if combined transaction taxes + brokerage exceeds alpha expectation.
alpha_expected > transaction.estimated_frictionBacktest Performance
Empirical walk-forward returns vs Benchmark (Nifty 50)
Asset & Event Scanner
Real-time extraction grid monitoring regulatory releases and mapping approved derivatives targets.
| TICKER | SECTOR | FILING TYPE | NLP POLARITY | FIREWALL | TARGET EXECUTION CONTRACT |
|---|---|---|---|---|---|
| RELIANCE | Energy | 10-Q | +0.78 | PASS | RELIANCE_FUT [LONG] |
| HDFCBANK | Banking & Finance | 10-K | +0.62 | PASS | HDFCBANK_FUT [LONG] |
| INFOSYS | IT & Tech | 8-K | -0.84 | HALT | BLOCKED_CONFIDENCE |
| TATAMOTORS | Auto & Manufacturing | 8-K | -0.45 | PASS | TATAMOTORS_950_PUT |
| TCS | IT & Tech | 10-Q | +0.12 | PASS | TCS_FUT [LONG] |
| ICICIBANK | Banking & Finance | 10-Q | -0.68 | PASS | ICICIBANK_1120_PUT |
| WIPRO | IT & Tech | 10-K | -0.22 | HALT | BLOCKED_CONFIDENCE |
| MARUTI | Auto & Manufacturing | 10-Q | +0.55 | PASS | MARUTI_FUT [LONG] |
API Sandbox
Connect your trading models in minutes. Designed for quantitative developers.
curl -X GET "https://api.candlery.in/v1/signals?ticker=RELIANCE" \
-H "Authorization: Bearer sk_live_8f7b...9e" \
-H "Accept: application/json"Click "Execute" above to send the handshake request and inspect the response payload.
System Telemetry
Active tracking of pipeline metrics, average latency, and API node operational status.
Our first system lost money. We published the audit.
Most AI startups hide their failures. We documented ours in the "Null Result" paper. We rebuilt the infrastructure from scratch to enforce radical quantitative honesty.
Read the Null Result Research"token-keyword">def validate_signal(signal_data):
"""
Ensures signal passes the Null Result criteria.
"""
"token-keyword">if signal_data.confidence < 0.95:
"token-keyword">raise HallucinationError("Confidence too low")
"token-keyword">if "token-keyword">not verify_historical_precedence(signal_data):
"token-keyword">raise PrecedenceError("No historical analog")
"token-keyword">return execution_compiler.build(signal_data)