CLOSED
AI & Methodology
MarketRipple is an AI-powered market intelligence platform for Indian investors. It connects market events to sectors, companies, historical patterns, and opportunities using real market data, evidence, and structured intelligence.
Companies Tracked
512
Curated NSE-listed universe
Historical Events
24
Verified events, 2008–2024
Relationship Types
7
Causal edge categories
Cascade Depth
4 levels
Upstream to downstream
The Workflow
At a Glance
| System | Input | Processing | Output |
|---|---|---|---|
| AI Search | User query + market data | Entity/intent matching → knowledge graph retrieval → AI synthesis | Sourced answer |
| Ripple Engine | Events + relationship graph | Multi-level dependency propagation, confidence attenuated per hop | Impact cascade |
| Opportunity Radar | Events + companies + sectors | Signal-density scoring engine | Opportunity score (0–99) |
| Confidence Engine | Evidence signals (sources, market/sector confirmation, precedent) | 8-factor point-based scoring | Confidence score (0–100) |
| AI Newsroom | Events + companies + real price data | Generation + deterministic fact-grounding validation | Published article |
AI Search
A natural-language question is matched against MarketRipple's own real event, company, and sector data — through entity and intent matching, not keyword search or a generic language-model response — and an AI model synthesises a sourced answer from what's retrieved.
Entities (companies, sectors, events, time periods) and query intent (comparison, trend, causal) are extracted from the question.
Extracted entities are matched against MarketRipple's knowledge graph of real event-company-sector relationships to find relevant events, companies, and sectors.
Supporting evidence is gathered from the matched events, sources, and any relevant historical precedent.
The retrieved evidence is synthesised into an answer grounded in that real data — not a generic model response.
Ripple Engine
Each relationship in MarketRipple's knowledge graph carries a stored confidence value. When an event's effects propagate through the graph, that confidence is attenuated at each hop — so a distant, indirect effect is shown with visibly lower confidence than a direct one.
MarketRipple maintains a knowledge graph with nodes for events, companies, sectors, commodities, and currencies. Edges represent 7 real relationship types, each carrying a stored confidence value.
New events are matched against real, verified historical market events (2008–2024) and existing graph structure to identify which relationships are economically meaningful.
Effects propagate through the dependency graph up to 4 levels deep. Each hop's confidence is attenuated by that edge's own confidence value, so distant, indirect effects are shown with visibly lower confidence than direct ones.
New events trigger graph re-evaluation on MarketRipple's regular processing cycle — not instantaneously, but on the same schedule described below in "How frequently is intelligence updated?"
Opportunity Radar
From a transparent, count-based formula — a base score plus credit for the number of corroborating events, and the breadth of companies and sectors involved in a developing situation, capped at 99. It's a real signal-density measure, not a fabricated number, and not a guaranteed return.
Event Impact
The number of real, corroborating events feeding into a developing situation — more independently confirming events raise the score.
Company Breadth
How many real companies are identified as exposed to the situation.
Sector Breadth
How many sectors the situation spans — a multi-sector theme scores higher than a single, narrow one.
Weights are versioned and retuned as the formula improves — see the factors above for what consistently matters, not a fixed public percentage split.
An opportunity score is an intelligence signal, not a guaranteed return or investment recommendation.
Explore Opportunity RadarConfidence Engine
A point-based sum across 8 real evidence signals — source count, historical precedent, market and sector confirmation, macro alignment, company sensitivity, AI self-certainty, and a volatility adjustment — capped at 100. It is computed, not an AI self-rating alone.
Source Count
0–15 pts
Distinct trusted sources behind the claim. More independent sources confirming the same fact raises confidence.
Historical Precedent
0–25 pts — the single largest factor
How many similar events exist in MarketRipple's library of 24 verified historical events (2008–2024), and how accurate past reads of those events turned out to be.
Market Confirmation
0–20 pts
Whether sectors or indices are already moving in the expected direction — real-time confirmation from the market itself.
Sector Confirmation
0–15 pts
Whether sector peers are independently confirming the same trend, beyond the specific company or event analysed.
Macro Alignment
0–15 pts
Whether the read is consistent with the prevailing macro regime — rates, inflation, growth — rather than fighting it.
Company Sensitivity
0–10 pts
How directly and predictably this company's fundamentals respond to this event type.
AI Self-Certainty
0–10 pts
The model's own calibrated certainty rating — one signal among many, never allowed to dominate the score alone.
Volatility Adjustment
−10 to +5 pts
A penalty in turbulent markets, a small bonus in calm ones — the same evidence supports a firmer conclusion when conditions are stable.
Points across all eight signals are summed and capped at 100 — a genuinely well-corroborated read comfortably clears the top of the range.
Validation
Before generating a company-impact analysis, real per-company price-move data is fetched and fed directly into the generation prompt. Three deterministic checks — no LLM involved — then run against the output before it can publish.
Flags the same boilerplate causal reason reused near word-for-word (≥90% text match) across two different companies in the same article.
Checks each company's stated impact tag against its real fetched price move. A ±2% or larger move tagged "neutral," or a move opposite the stated tag, is flagged.
Compares the article's language against the source event's own language to catch a draft regulation described as finalized, or vice versa.
Currently in shadow mode: violations are logged for review, not yet blocking publication, while MarketRipple observes real-world violation rates before turning this into a hard publish gate.
Fact
What the source or data indicates — a real price move, a real filing, a real historical outcome.
AI Interpretation
MarketRipple's model-generated read of what the evidence means, clearly labelled as such.
Prediction / Outlook
A forward-looking view — always probabilistic, never presented as certain or guaranteed.
Data
Only real, verifiable sources — never a source claimed without a real connection behind it.
Live equity prices, index levels
NSE/BSE filings, RBI, SEBI, PIB
Economic Times, Moneycontrol, Business Standard, Livemint, NDTV Profit
24 verified events, 2008–2024
Multi-provider LLM fallback chain
The Ripple graph, opportunity and confidence scores
Honest Disclosure
Transparency requires honesty about what AI can and cannot do. These are the genuine limitations of MarketRipple's analytical systems.
MarketRipple analyses probabilities and historical patterns. It cannot predict market movements with certainty. Treat high-confidence signals as strong hypotheses, not facts.
Analysis is based exclusively on publicly available information — filings, market data, news, and official announcements. MarketRipple has no access to private or insider information.
Historical patterns are the foundation of confidence scoring. In genuinely unprecedented events, confidence scores will be lower — as they should be.
Confidence percentages are calibrated uncertainty, not statistical guarantees. A 75% confidence signal will be wrong roughly 25% of the time.
News and exchange sources are polled on a fixed schedule, not streamed live. For intraday trading decisions, always verify against primary sources.
Does
Does Not
No. MarketRipple is designed to augment human investment judgment, not replace it. We recommend using MarketRipple to generate and stress-test hypotheses, then verifying conclusions against primary sources (BSE/NSE filings, RBI releases, company annual reports) before making any investment decision.
MarketRipple does not provide personalised investment advice. All analysis is for informational purposes only. Past patterns do not guarantee future outcomes.
Questions
A natural-language question is matched against MarketRipple's own event, company, and sector data through entity and intent matching — not vector embeddings — and an AI model synthesises a sourced answer from what's retrieved. Answers cite the underlying events.
Events are matched to companies either directly (the company is named in the event) or through the Ripple Engine's relationship graph, which connects an event's sector or commodity to a company through one of 7 defined relationship types.
Each relationship (edge) in the graph carries a stored confidence value. When an event's effects propagate through the graph — up to 4 levels deep — that confidence is multiplicatively attenuated at each hop, so distant effects are shown with visibly lower confidence than direct ones.
From a transparent, count-based formula: a base score plus credit for the number of corroborating events, the breadth of companies involved, and the breadth of sectors involved, capped at 99. It's a real signal-density measure, not a guaranteed return.
A point-based sum across 8 real evidence signals — source count, historical precedent (the largest factor), market confirmation, sector confirmation, macro alignment, company sensitivity, AI self-certainty, and a volatility adjustment — capped at 100.
Yes — MarketRipple compares current events against a set of 24 verified historical market events (2008–2024) where a genuinely similar precedent exists. This is the single largest factor in the confidence score, but it provides context, not a guarantee of future returns.
Company-impact analysis is grounded in real fetched price data before generation, and deterministic (non-AI) validators check every draft afterward for shared boilerplate reasons, sentiment tags that contradict the real price move, and status mismatches — before anything publishes.
No. MarketRipple is a market research and intelligence tool. It does not replace the personalised advice of a SEBI-registered investment advisor, and all investment decisions remain the user's responsibility.
News and exchange sources are polled every 15 minutes, and regulatory sources hourly. Article generation runs on a 5-minute cycle over newly classified events. This is a scheduled cadence, not literal real-time streaming.
Every AI Newsroom article carries an Evidence section split into Fact — what happened, sources, historical outcomes — and AI Interpretation — the model's own read, clearly labelled as such.
Continue Exploring
Explore how MarketRipple reasons through real market events — and see the pipeline behind it.