Markets are complex.
Intelligence should be collaborative.
Dhanvi is building an AI-native investment intelligence system where specialized agents research markets, develop competing strategies, evaluate risk, and learn from outcomes.
Works alongside institutional workflows. Governed by deterministic risk gates.
Institutional asset management was designed around human bottlenecks.
Specialized teams traditionally operate in separate functional silos — passing memos, spreadsheets, and delayed models across organizational boundaries with inevitable friction.
Siloed analysis & sequential latency
When research is fragmented, macro assumptions diverge from company models, and trade ideas wait for weekly allocation committee meetings.
- Cognitive limits: Human analysts cannot parse thousands of global SEC filings, earnings calls, and tick streams in parallel.
- Context loss: When positions are closed, the original hypothesis, market regime context, and attribution details are rarely codified.
- Reactive risk: Human risk committees frequently adjust exposures after market volatility spikes have already occurred.
Software-native collaborative intelligence
Instead of isolated teams waiting on quarterly memos, specialized autonomous agents interrogate data streams concurrently, synthesize consensus, and formulate testable strategies.
Specialized intelligence.
One shared understanding.
Dhanvi approaches markets through specialized research agents, each focused on a different dimension of the investment environment.
News Intelligence
Monitors relevant financial news, company developments, filings, macro events and geopolitical developments to identify potentially market-relevant information.
- • Financial news & wires
- • Company filings (8-K, 10-Q)
- • Earnings releases & transcripts
- • Global policy & geopolitical events
- • Relevance scoring & tagging
- • Entity recognition & linkage
- • Context extraction
- • Potential market impact evaluation
- • Structured events
- • Research context dossier
- • Downstream strategy signals
From information to decision.
How market observations move through multi-agent research, hypothesis debate, portfolio allocation, independent risk verification, and institutional memory.
Observe
Receives market microstructure, global regulatory filings, tick streams, macro indicators, and corporate disclosures in parallel.
Many hypotheses.
One portfolio.
Dhanvi simulates multiple independent strategies simultaneously. Competing alphas are continuously evaluated and dynamically weighted across shifting market regimes.
Orderly equity trends, normal credit spreads, and active market liquidity.
A system that remembers
why decisions were made.
In human organizations, institutional memory dissolves when key people leave. Dhanvi preserves the complete forensic context behind every hypothesis, risk calculation, and trade — creating a permanent record of what was believed, why it happened, and what was learned.
When semiconductor supply-chain bottleneck constraints and rate sensitivity re-emerged 6 months later, Dhanvi’s vector retrieval surfaced this exact decision trace with 92% semantic alignment — preventing recency bias and accelerating parameter calibration.
We don't just describe the idea.
We test it.
Dhanvi Experiments documents prototypes, simulations and research conducted while building the system — including what was tested, how the system behaved, what happened and what we learned.
Dhanvi Experiments is an expanding research log. More experiments will be published here as Dhanvi develops.
Multi-Agent Paper Trading
Testing whether specialized AI agents could coordinate across research, analysis and simulated trading during a complete market session.
EMA Momentum Reversal Backtest
A controlled historical backtest of the DHA-DEMO-EMA-MOM strategy on RELIANCE.NS, testing execution behavior, costs and strategy performance across 72 daily bars.
Early Strategy Generation Backtest
The significance wasn't the number. It was what came next.
Evaluation over execution.
The primary architectural challenge was validating that the system could reject low-conviction ideas rather than attempting to trade every candidate identified in the morning stream.
How intelligence moved through the system.
Five specialized agents operated in sequence during the experimental session. Click or hover any agent to inspect its specific task.
From pre-market ingestion to session close.
The continuous multi-agent cycle operated across the complete market hours between 08:45 and 15:15.
System initialization and macro baseline calibration prior to open.
News Agent processes market open bulletins and filing disclosures.
Converts unstructured information into broader market research context.
Evaluates intraday conditions and identifies 459 candidate opportunities.
Conducts deeper company-level screening on shortlisted candidates.
Determines the 33 opportunities progressing to simulated paper trades.
Market session close. Ledger recorded and session metrics finalized.
This was a single paper-trading experiment conducted during development of Dhanvi. It does not represent live trading performance or expected future returns. Paper trading and simulations may differ materially from real trading due to factors including execution, fees, spreads, slippage, latency, liquidity and market impact. Past or simulated performance does not guarantee future results.
What we were testing
What Experiment 001 does not establish
A single paper-trading session does not establish:
- expected future returns
- long-term profitability
- live trading performance
- strategy robustness across market regimes
- realistic execution performance
- scalability to large amounts of capital
The research continues.
Experiment 001 tested multi-agent coordination. Experiment 002 tested a single strategy inside our backtest engine. Experiment 003 explored early algorithmic strategy generation. More experiments will be published here as Dhanvi develops.
Autonomous intelligence does not mean
unrestricted control.
In financial systems, machine learning models cannot be allowed to alter their own safety boundaries. Dhanvi isolates risk management in a separate, deterministic software layer outside agent reasoning — enforcing hard rules that AI cannot override or negotiate away.
Position Limits
Prevents single-name idiosyncratic exposure from disproportionately impacting total portfolio value.
Exposure Bounds
Enforces strict systemic limits on leverage and directional market beta, regardless of agent conviction.
Drawdown Protections
Deterministic automated de-risking when trailing portfolio drawdown touches predetermined bounds.
Liquidity Capacity
Ensures no transaction creates market impact beyond strict thresholds; mandates orderly liquidations.
Execution Safeguards
Locks out execution during uncharacteristic spread widening or extreme microstructure illiquidity.
Human Oversight
Requires human sign-off for large re-allocations or non-standard asset onboarding outside normal bands.
Hardware Kill Switch
An independent manual and automated kill switch that cancels pending orders and flattens positions immediately.
Three product directions.
Dhanvi is designing software-native financial intelligence for different scales of deployment — from institutional quantitative infrastructure to conversational research APIs and future personal tools.
Dhanvi Institutional
Modular multi-agent deployment designed for quantitative desks, family offices, and institutional asset managers requiring autonomous research with deterministic risk governance.
- —Research intelligence pipelines
- —Strategy hypothesis generation & backtesting
- —Portfolio covariance analytics
- —Deterministic risk intelligence
- —Immutable decision memory engine
Dhanvi Intelligence
A structured query interface allowing analysts to interrogate multi-agent findings, cross-asset regimes, and corporate disclosures in real time.
- —Natural-language research queries
- —Historical context & analog retrieval
- —Cross-asset market relationships
- —Decision intelligence breakdowns
Dhanvi Personal
A strictly non-custodial research assistant helping self-directed investors evaluate portfolio factor health and market regime shifts.
- —Personal portfolio factor research
- —Portfolio health & regime insights
- —Scenario stress-testing
- —Simulated paper-research laboratory
Real-time research telemetry.
Inspect the telemetry stream as specialized agents formulate hypotheses, evaluate risk boundaries, and archive decision context into memory.
The investment institution of the future may look very different.
Dhanvi is being built at the intersection of artificial intelligence, financial research, quantitative systems, and autonomous software. Join our early-access waitlist to participate in research updates and private previews.