DroxenBot: A Real-Time Autonomous Agent for On-Chain Detection and Ranking of Emerging Digital Assets
Author: Abdullahi Labaran
Background: B.Tech Computer Engineering (AI & Data Science)
Research Interests: Autonomous Systems, Applied Machine Learning, AI Engineering, Modeling & Simulation
Email: baffahlabaran01@gmail.com
LinkedIn: www.linkedin.com/in/abdullahi-labaran
Year: 2026
Research repository: This public repository documents the research design, system architecture, methodology, and experimental evolution of DroxenBot. The production implementation is maintained separately from this repository.
DroxenBot is an experimental autonomous market-intelligence agent designed to monitor and rank emerging digital assets using real-time on-chain activity, decentralized-exchange market data, and behavioral wallet signals.
The system combines blockchain data streams, decentralized-exchange analytics, rule-based risk and eligibility filtering, and heuristic scoring to prioritize newly launched assets for continuous monitoring.
The project investigates how autonomous decision-support systems can operate in noisy, high-velocity financial environments where newly launched assets have limited historical information available at detection time.
Newly launched decentralized assets can emerge and begin trading within short periods, creating a rapidly changing environment for continuous monitoring.
Compared with established financial assets, newly launched tokens may have limited historical information available when they first become observable. Early-stage monitoring therefore requires systems capable of:
- Continuous data collection
- Feature extraction from heterogeneous signals
- Risk and eligibility filtering
- Candidate prioritization
- Real-time alert generation
- Post-detection monitoring
DroxenBot investigates how a rule-based autonomous monitoring system can combine these processes into a continuous decision-support pipeline.
The project investigates:
- Can algorithmic filtering help identify potentially significant emerging assets at an early stage of their market activity?
- Which on-chain and market signals are useful for prioritizing newly launched assets for monitoring?
- Can wallet-behavior signals provide useful information for early-stage asset monitoring?
- How can autonomous monitoring systems reduce noise when evaluating rapidly changing decentralized markets?
The project explores:
- A real-time autonomous monitoring pipeline for newly launched decentralized assets
- A multi-stage architecture combining data collection, preprocessing, risk screening, scoring, classification, storage, and alert delivery
- Integration of heterogeneous market, transaction, liquidity, and wallet-related signals
- Rule-based risk and eligibility filtering prior to candidate scoring
- Heuristic scoring and tier classification for monitoring prioritization
- Post-detection tracking of selected assets and growth milestones
- Iterative system development based on observations collected during live deployment
The reported performance observations are discussed as observational findings rather than controlled predictive-performance estimates. The accompanying research paper documents the measurement limitations and proposes a more rigorous evaluation protocol.
DroxenBot is organized as a staged monitoring pipeline:
flowchart TD
A["DATA SOURCES<br/><br/>Solana on-chain data<br/>DEX market data<br/>Token metadata<br/>Wallet / transaction data"]
B["DATA COLLECTION<br/><br/>Token discovery<br/>Market data collection<br/>Liquidity data<br/>Transaction data"]
C["PREPROCESSING & FEATURE EXTRACTION<br/><br/>Normalization<br/>Market features<br/>Transaction features<br/>Wallet and risk signals"]
D["RISK & ELIGIBILITY FILTER<br/><br/>Liquidity checks<br/>Ownership / concentration<br/>Bundle / behavior checks<br/>Market thresholds<br/>Anti-wash rules"]
E["SCORING & CLASSIFICATION<br/><br/>Heuristic scoring<br/>Momentum signals<br/>Trend detection<br/>Tier classification"]
F["STORAGE & MONITORING<br/><br/>PostgreSQL<br/>Redis<br/>Token and price history"]
G["ALERT & OUTPUT<br/><br/>Telegram alerts<br/>Ranked signals<br/>Monitoring results"]
A --> B
B --> C
C --> D
D --> E
E --> F
E --> G
The architecture separates data acquisition, feature construction, risk screening, candidate evaluation, state management, monitoring, and alert delivery within a continuous monitoring pipeline.
The system operates on heterogeneous inputs including:
- Solana on-chain data
- Decentralized-exchange market data
- Token metadata
- Wallet and transaction information
The collection stage discovers candidate assets and gathers:
- Market data
- Liquidity information
- Transaction activity
- Relevant wallet-related signals
Raw observations are transformed into structured signals including:
- Market features
- Transaction features
- Liquidity-related features
- Wallet and behavioral signals
- Time-based features
Rule-based checks remove candidates that fail configured criteria involving:
- Liquidity
- Ownership and concentration
- Behavioral or bundle-related signals
- Market thresholds
- Anti-wash-trading rules
These checks are intended for operational risk screening and noise reduction. They do not guarantee that an asset is safe, legitimate, or profitable.
Candidates that pass the filtering stage are evaluated using a weighted heuristic scoring process.
The resulting score is used to:
- Prioritize candidates
- Identify momentum-related signals
- Detect trends
- Assign monitoring tiers
The tiers are intended for prioritization and are not calibrated probabilities of future price appreciation.
The system maintains persistent and frequently accessed state using:
- PostgreSQL
- Redis
- Token and price history
Selected assets can subsequently be monitored for changes in market activity and predefined growth milestones.
Candidates meeting configured conditions can generate:
- Telegram alerts
- Ranked monitoring signals
- Post-detection monitoring results
The system was iteratively developed and evaluated during an approximately nine-month research period.
| Phase | Approx. Signals / Day | Observed Hit Rate |
|---|---|---|
| Early prototype | ~80 | ~10–20% |
| Optimized system | ~10–15 | ~70–85% |
Important measurement note: These hit-rate figures are observational estimates from the project's deployment period. They were not produced using a single fixed success threshold applied consistently across all stages, nor were they evaluated against a controlled baseline. Consequently, they should not be interpreted as standard precision, recall, accuracy, or validated predictive performance.
The accompanying research paper provides the measurement methodology, limitations, and proposed evaluation protocol for obtaining more rigorous performance estimates.
During the research period, DroxenBot was deployed as a Telegram-based monitoring system and operated under live market conditions.
The deployment period involved:
- Approximately nine months of iterative development and monitoring
- Large-scale observation of newly launched assets
- Automated Telegram alert delivery
- Post-detection tracking of selected assets
- Continuous refinement of filtering and scoring rules
The production implementation is maintained separately from this public research repository.
| Component | Technology |
|---|---|
| Language | Python |
| Backend | FastAPI |
| Database | PostgreSQL |
| Cache | Redis |
| Blockchain Data | Helius API |
| Market Data | DEX analytics APIs |
| Messaging | Telegram Bot API |
| Deployment | Railway / Render |
This public repository contains research documentation and architecture materials rather than the complete production implementation.
DroxenBot-Research/
│
├── docs/
│ └── architecture.svg
│
├── src/
│ └── README.md
│
├── README.md
├── LICENSE
└── requirements.txt
The full production system is maintained separately. The public repository is intended to provide a reproducible description of the research design, architecture, methodology, and experimental evolution without exposing the complete production implementation or proprietary filtering configuration.
The system evaluates candidates using heterogeneous signals such as:
- Liquidity depth
- Buy/sell pressure
- Volume activity
- Holder concentration
- Wallet activity
- Time since launch
- Market-capitalization momentum
- Transaction activity
These signals are combined after preprocessing and risk/eligibility filtering.
DroxenBot uses a weighted heuristic scoring approach combining:
Market metrics + on-chain activity + behavioral indicators
The resulting score is used for candidate prioritization and monitoring-tier assignment.
The current system is intentionally heuristic rather than a trained predictive machine-learning model. Future work will evaluate learned ranking approaches against the existing heuristic system.
This project connects to:
- Autonomous Agents
- Real-Time Systems
- Applied Machine Learning
- Modeling & Simulation
- Data-Driven Decision Support
- Blockchain Analytics
Observations during the deployment period included:
- Iterative filtering was associated with a reduction in the number of daily signals and an increase in the project's observed hit-rate estimates.
- Wallet and behavioral activity was frequently observed alongside assets that subsequently experienced substantial market-capitalization growth.
- Newly launched assets exhibited rapidly changing market and transaction signals that motivated continuous rather than purely historical analysis.
These are observational findings from live deployment. They do not establish causality, statistical significance, or generalizable predictive performance.
The current study has several limitations:
- No controlled randomized baseline
- No consistently applied success threshold across the full development period
- No formal precision, recall, F1, or calibration analysis for the reported historical hit-rate estimates
- Potential selection and survivorship effects
- Changing market conditions across the development period
- Heuristic thresholds were iteratively refined during development
- The current evaluation does not establish causal relationships between individual signals and subsequent market movements
These limitations motivate the evaluation framework proposed for future work.
-
Rigorous Evaluation Protocol Apply a fixed success criterion consistently across detections and report appropriate classification and ranking metrics.
-
Baseline Comparisons Compare the multi-stage system against random selection and simple single-metric filtering strategies.
-
Machine-Learning Ranking Train supervised or semi-supervised ranking models using accumulated detection history and compare them with the current heuristic scorer.
-
Temporal and Graph-Based Modeling Represent wallet and transaction relationships as temporal graphs for behavioral analysis.
-
Simulation and Backtesting Replay historical market streams under controlled experimental conditions.
-
Cross-Chain Generalization Evaluate whether the monitoring framework can be adapted to additional blockchain ecosystems.
-
Risk-Aware Decision Support Incorporate uncertainty estimates and explicit liquidity constraints into candidate prioritization.
-
Autonomous Execution as a Separate Research Problem Evaluate automated execution independently under controlled risk limits rather than conflating execution performance with monitoring performance.
The full research paper documents the system architecture, methodology, related work, observational findings, limitations, and proposed evaluation framework.
Preprint: To be added after arXiv publication.
DroxenBot is a research and educational project. Nothing in this repository constitutes financial, investment, or trading advice.
Digital-asset markets are highly speculative and can involve substantial financial loss. Historical observations from the project should not be interpreted as guarantees of future performance.
If you reference this research, please cite the associated preprint after publication:
A. Labaran, "DroxenBot: A Real-Time Autonomous Agent for On-Chain Detection and Ranking of Emerging Digital Assets," 2026.