International Innovation Challenge

2.0
September 08-09, 2025
Release of problem statements

Venue: Smt. Vasanti R Pai Auditorium

Date: September 08, 2025

1. lmage based breed recognition for cattle and buffaloes of India

2. AI based development of Laser based QR Code marking on 'track fittings on Indian
Railways'.
3. Development of Sensor for Detection Of Microplastics
4. Automated Compliance Checker for Legal Metrology Declarations on E-Commerce
Platforms
5. Smart Tourist Safety Monitoring & Incident Response System using Al, Geo-Fencing,
and Blockchain-based Digital ID
6. Develop a blockchain-based system for botanical traceability of Ayurvedic herbs,
including geo-tagging from the point of collection (farmers/wild collectors) to the final
Ayurvedic formulation label.
7. Develop API code to integrate NAMASTE and or the International Classification of
Diseases (ICD-11) via the Traditional Medicine Module 2 (TM2) into existing EMR
systems that comply with Electronic Health Record (EHR) Standards for India.
8. Intelligent Pesticide Sprinkling System Determined by the Infection Level of a Plant
9. Low-Cost smart transportation solution for Agri produce from remote farms to nearest
motorable road in NER Region
10. Low-Cost smart transportation solution for Agri produce from remote farms to nearest
motorable road in NER Region
11. Gamified Environmental Education Platform for Schools and Colleges
12. Smart Community Health Monitoring and Early Warning System for Water-Borne
Diseases in Rural Northeast India
13. Digital Platform for Centralized Alumni Data Management and Engagement
14. Al-Based Internship Recommendation Engine for PM Internship Scheme
15. Smart Waste Segregation and Recycling System
16. Improved Onion storage technology for enhancing shelf life of onions
17. Designing and development of an application for on spot assessment of Roof Top Rain
water harvesting and artificial recharge potential and size of the RTRWH and AR.
18. AI-Based Farmer Query Support and Advisory System
19. To design and develop an indigenous, secure, and ruggedized Customer Premises
Equipment (CPE) system to enable encrypted and reliable internet connectivity onboard
 Indian naval ships under the Secure Internet for onboard Naval Ships (SIFON)
framework in coordination with TSP.
20. Design AI-powered bots for small farmers to aid with soil testing, pest detection, crop
management, irrigation, and accessing financial aids.
21. Develop solutions for tracking immunization services, monitoring temperature-
sensitive goods for vaccines and medicines during transport, and creating apps for
disease outbreak management.
22. Develop a mobile application for tracking district officials' field visits and inspections
using geo-tagging and digital authentication.
23. Develop innovative assistive technologies and inclusive infrastructure to empower
visually impaired individuals with enhanced navigation, mobility, and independence,
thereby improving their social interaction and overall quality of life.
24. Develop a solution for connecting farmers with local markets, enabling real-time price
discovery, efficient supply chain management, and fair pricing for agricultural produce.
25. Create a user-friendly software solution to map government schemes to beneficiaries
based on their socio-economic background? Ensure real-time access and a citizen-
centric design for efficient grievance redressal.
26. develop a unique and efficient mechanism to automatically stop the transfer of social
security funds to deceased under various schemes? The solution should ensure timely
identification of deceased beneficiaries, leveraging integration with death registration
databases, Aadhaar authentication, or periodic life certification through digital methods
to prevent misuse and save government funds.
27. Create accessible VR and AR-based tools to enhance hands-on, industry-relevant
training for students.
28. Develop an integrated "All in one" mobile app for MUJ including smart parking system
with pre-notification and navigation to vacant parking space, smart and easy navigation
within MUJ campus, checking availability of MUJ employees with location and phone
number, alumni connection, complaints and tracking for general services.
29. AI-Driven Public Health Chatbot for Disease Awareness
Description
Create a multilingual AI chatbot to educate rural and semi-urban populations about
preventive healthcare, disease symptoms, and vaccination schedules. The chatbot should
integrate with government health databases and provide real-time alerts for outbreaks.
Expected Outcome
A chatbot accessible via WhatsApp or SMS, reaching 80% accuracy in answering health
queries and increasing awareness by 20% in target communities.
Technical Feasibility
Built using NLP frameworks (e.g., Rasa, Dialogflow) with APIs for health data integration,
deployable on cloud platforms for scalability.
30. Quality Control for Medical Imaging Studies
Description
Build an AI-powered pipeline that automatically checks CT scans for resolution,
completeness, and anonymization compliance before they are used for training AI models.
Expected Outcome
 A quality-check dashboard flagging failed studies with reasons (low resolution, missing
slices, PHI leakage) and achieving >90% accuracy in anomaly detection.
Technical Feasibility
Leverage PyTorch/TensorFlow models for resolution detection, regex + NLP for report
validation, integrated into a Databricks/Azure workflow.
31. Multi-Label Chest X-Ray Findings Detection
Description
Most X-ray images contain multiple findings (e.g., pneumonia + effusion). Create an AI
system that can detect multiple abnormalities simultaneously rather than just
normal/abnormal.
Expected Outcome
A model that outputs probabilities for at least 3–5 disease categories per X-ray, with a
dashboard to visualize predictions and confidence scores.
Technical Feasibility
Public datasets (NIH ChestX-ray14, CheXpert) can be used. Pretrained models
(ResNet/DenseNet) can be fine-tuned within 36 hours. Grad-CAM visualization can be
added for interpretability.
32. Real-Time Imaging Workflow Orchestration
Description
In hospitals, multiple scans arrive every minute. Design a workflow tool that ingests
DICOM files, routes them through anomaly detection models, and pushes prioritized
cases to a mock “radiologist dashboard.”
Expected Outcome
An end-to-end pipeline simulation: input (DICOM/PNG) → AI model → results sorted
by urgency. A simple web dashboard shows the case queue with timestamps.
Technical Feasibility
Can be simulated using Flask/Django + pre-trained models. Use open-source DICOM
libraries (pydicom). Teams focus on workflow automation + AI integration.
33. Privacy-Preserving Medical Image Sharing
Description
Medical images often contain sensitive information. Build a pipeline that automatically
removes text labels and anonymizes metadata while preserving diagnostic quality.
Expected Outcome
A prototype that takes in medical images with patient names/IDs and outputs
anonymized versions, along with a report comparing original vs anonymized images.
Technical Feasibility
Use OCR (Tesseract/EasyOCR) for text detection and image inpainting. Remove
DICOM metadata fields using pydicom.
34. AI-Powered Multi-Modal Classroom Engagement Analyzer (Non- Healthcare)
Description
Build an AI tool that analyzes classroom videos and determines if students are engaged,
distracted, or absent using both facial expressions (vision) and voice/text cues
(audio/NLP).
 Expected Outcome
A demo that labels engagement levels of students in a short clip and a dashboard with
engagement statistics over time.
Technical Feasibility
Open datasets: AffectNet (facial expressions), RAVDESS (speech emotion). Use
pretrained emotion detection + late-stage fusion for multi-modal analysis.
35. AI-Powered Spine Posture s Curve Estimation from Human Body
Description
Create a computer-vision system that estimates spine alignment and curvature from
normal RGB/depth images (webcam or phone camera). The system should identify
body keypoints (shoulders, hips, back) and approximate spinal curves to detect
abnormal postures, scoliosis tendencies, or long-term musculoskeletal risks.
Expected Outcome
A real-time posture monitoring tool that achieves >85% accuracy in estimating spinal
alignment compared to physiotherapist assessments, and provides visual feedback for
correction (e.g., “leaning left”, “slouched posture”).
Technical Feasibility
Use pose-estimation frameworks (MediaPipe Pose, OpenPose, or SMPL model fitting)
to extract 2D/3D skeletal landmarks. Apply curve-fitting (splines/polynomials) to
approximate the spine’s natural curve and compute posture deviation metrics.
Deployable as a desktop/mobile app for physiotherapy, ergonomics, or fitness use
cases.
36. Dashboard development using AI-ML techniques to interpolate tropospheric
Precipitable Water (PW) content using zenith-wet delay from GNSS observations
Background:
Troposphere is a critical layer of earth’s atmosphere which influences weather patterns.
Troposphere holds Precipitable Water (PW) and affects the transmission of GNSS
signals. GNSS reference stations capture signals from multiple satellites and have
observations, which can help estimate tropospheric wet content. If the precise
coordinates of reference station and satellites are known, the GNSS reference station
can compute zenith-wet delay. The zenith-wet delay along line-of-transmission of
GNSS signal can help understand the distribution of PW in a given area and help
forecast rainfall weather alerts.
Detailed Description:
In this task, the zenith-wet delay, observed over 24 hours to 7 days, at multiple GNSS
stations shall be provided. Additionally, the elevation and azimuth of GNSS satellites
shall be provided. The challenge is to develop an AI/ML-powered algorithm to
interpolate tropospheric wet delay over a given region of GNSS network. The software
solution shall
• Accept zenith-wet delay observations from multiple stations
• Compute temporal wet delay between multiple stations based on elevation and
azimuth of GNSS satellites
• Apply AI-ML techniques to improve interpolation accuracy
 • Display results on a dashboard
Expected Solution:
The developed solution shall provide:
• A dashboard which shall show the interpolated wet delay at any point in the given
area
• Accept third-party observations of wet-delay and compute error in the interpolation
results
• Extend the algorithm to extrapolate wet delay in regions beyond the given GNSS
network
37. Ultra-Extreme Image Compression for Low-Bandwidth Transmission Using AI-
Based Super-Resolution (SR)
Background
Super-resolution is a technique in computer vision that enhances the resolution and
quality of low-resolution images or videos using deep learning. It reconstructs fine
details, sharpens edges, and improves clarity. This technology is crucial for applications
like medical imaging, satellite imagery, surveillance, and digital restoration where
detail preservation is critical.
Detailed Description:
Develop an AI-powered image compression and reconstruction system that does
following:
• Input: Takes a high-resolution image (~1 MB).
• Compression: Reduces it to a maximum of 1 KB for transmission or storage.
• Reconstruction: Recovers a high-resolution image (~1 MB) from the 1 KB file with
minimal perceptual distortion.
• Output Quality: The reconstructed image should be visually close to the original,
preserving key structures, colours, and features, while keeping Mean Opinion
Score (MOS) and perceptual similarity metrics (LPIPS, MS-SSIM) within
acceptable thresholds.

Expected Solution:

• Encoder: Converts input image into a ≤1 KB compressed representation.

• Decoder: AI model that reconstructs the image at full resolution.
• Performance Report:
• Compression ratio achieved.
• MS-SSIM, LPIPS, and MOS scores vs. baselines (JPEG, AVIF at same size).
• Demonstration Application: Web or desktop app showing side-by-side comparison of
original vs. reconstructed images. Any 1MB image can be taken
 38. Cross Modal SAR-Optical Image Matching/Registration
Background
SAR and optical imagery provide complementary information about the Earth's surface.
SAR is sensitive to surface roughness and dielectric properties, while optical imagery
provides information about surface reflectance and colour. Combining these modalities
offers improved capabilities for various applications including land cover classification,
change detection, disaster monitoring, and urban planning. However, directly matching
(Image Registration) these images is challenging due to differing sensor characteristics,
geometric distortions, and viewing geometries.
Detailed Description
The solution should explore and implement advanced feature extraction and matching
techniques to overcome these challenges and establish a strong correspondence
between SAR and Optical imagery. The proposed solution should implement following
aspects:

• Data Acquisition & Preprocessing: Download and preprocessing of relevant SAR

and optical image pairs from the Cloudsen12 dataset, including geometric
correction, co-registration, and radiometric calibration.

• Feature Extraction: Investigate and implement robust feature extraction techniques

suitable for both SAR and Optical images. This will include exploring both
conventional (e.g., SIFT, HOG, corner detectors) and deep learning-based feature
descriptors (e.g., SiLK, SuperPoint, LoFTR, ViT).

• Geometric Transformation Estimation: Estimate the geometric transformation

(e.g., affine, projective) between the matched features to align the SAR and optical
images.

• Performance Evaluation: Quantitatively evaluate the performance of the developed

algorithm using standard metrics such as:
➢ Matching Accuracy: Precision, Recall, F1-Score of matched features.
➢ Geometric Accuracy: Root Mean Square Error (RMSE) of control points.

• Comparative Analysis: Compare the performance of different developed conventional

& ML/DL based feature extraction and matching techniques.
Expected Solution

Development of a robust and accurate algorithm for matching Synthetic Aperture Radar
(SAR) and optical images, leveraging the rich and readily available Cloudsen12 dataset.
Dataset Link: [Link]
 39. AI/ML Optical Flow based Frame interpolation of Satellite Images for generation
of High Temporal Resolution data
Background:
Satellite images are often captured at fixed intervals (e.g., every 30 minutes for
geostationary satellites like INSAT, every 10 minutes for Himawari/GOES satellites or
every few days for polar-orbiting satellites). This leads to limitation in temporal
resolution that restricts near real-time monitoring of dynamic phenomena such as
cyclones, thunderstorms, floods, and rapid land changes. Traditional temporal
interpolation is limited with incorrect results (blurred images & other artefacts) and
they fail to capture fast-moving or non-linear changes in cloud dynamics.
This problem proposes to develop an AI/ML based Optical Flow frame interpolation
technique to generate intermediate frames between two consecutive satellite images to
effectively enhance temporal resolution, enabling more frequent observations without
additional satellite resources.
Detailed Description:
The developed software should include and accomplish the following steps:
• Develop an optical flow-based model to estimate pixel-wise motion vectors
between consecutive satellite frames.
• Generate synthetic intermediate frames using AI-driven frame interpolation
methods (e.g., Deep Optical Flow, RAFT, Super SloMo).
• Improve temporal resolution of satellite imagery (e.g., from 30 minutes to 15
minutes and further to 7.5 minutes interval or equivalent).
• Validate results against real higher temporal resolution datasets (like
Himawari/GOES-R) using image similarity metrics such as SSIM, MSE, PSNR,
FSIM, etc.
Expected Solution:

Solution for the challenge should include a prototype for AI-powered frame interpolation for
satellite imagery that should comprise of:
1. Frame Interpolation: Takes two consecutive satellite images and generates intermediate
frames. Should include:
a. Data collection of satellite datasets from online (INSAT-3D/3DR, Himawari-8,
GOES-R)
b. Optical Flow Estimation – can explore and use AI/ML models
c. Frame Interpolation – can explore and use interpolation networks for synthesis of
new frames
2. Visualization: Develop a dashboard showing original vs. interpolated satellite
animations. Should include:
a. After interpolation of a set of satellite frames, the output should be displayed with
time-lapse animations.
b. Animations to be present both for ground truth (original satellite frames) and also
for interpolated frames.
 3. Evaluation: Generate a report comparing results with ground truth (existing frames).
Should include:
a. Comparison of generated frames with the ground truth using metrics of SSIM, MSE,
FSIM, etc. Include plots of the results
b. Any other suitable metrics can also be explored and used. Metrics should be suitable
to capture the cloud movements.
Hence, the developed solution should have the capability of:
1. Improving temporal density of satellite data without new hardware.
2. A deployable pipeline (Python-based, using TensorFlow/PyTorch/any other framework
+ OpenCV).

40. Vision based Intelligent RPA agent

Background:
Most RPA systems today work by using fixed selectors (like XPath, control IDs, or
DOM paths). These tend to break the moment the interface changes even slightly —
which means a lot of manual re-configuration. Now with the developments done in the
field of vision transformer models and a number of powerful open-source vision-
language models, it has become feasible with reasonable resources to “look at” a screen
and understand what’s on it. This has led to development of visual RPA agents that can
interact with digital interfaces — whether it’s a website, desktop screen or a dashboard
— based purely on what they see and a natural language instruction.

Detailed Description:
In this challenge, students need to build a generic visual RPA agent that interacts with
digital interfaces using vision and natural language. Instead of relying on predefined
DOM rules or CSS selectors, the agent should:
• Take a screenshot of the current interface (web/desktop)
• Use an existing Vision Transformer model (e.g., DETR, DINO, Swin, SAM, etc.) to
detect actionable UI elements on the screen — like buttons, input fields, dropdowns,
links, etc. The existing models may be further fine-tuned for improved accuracy.
• Run an open-source OCR tool to extract the text from those elements.
• Read a natural language instruction from the user and match that instruction to the
correct visual element using a multi-modal model (e.g. CLIP, sentence transformers,
etc.).
• Perform the action using only open-source automation tools (e.g., Selenium,
Playwright, PyAutoGUI etc.).

RPA agent should work on Web applications and at least one of the operating systems
(Windows or Linux)
Only self-developed or open-source software should be used for the development. Even if
APIs to language models are used, the model behind that API should be open-source and
available for offline deployment.
Expected Solution:
The final submission should include:
• A perception module that identifies actionable UI elements using a ViT-based model
• An OCR component to read any visible text
• A matching module to link a natural language instruction with the correct on-screen
element
• An automation layer that performs the action on the real interface using open-source
tools
• A short demo showing the agent successfully completing a multi-step task (e.g., filling
and submitting a form, navigating a dashboard, downloading a report, etc.) based only
on natural language input.

41. Post Quantum Cryptography (PQC) based TLS Termination Reverse Proxy
Background
Transport Layer Security (TLS) is the backbone of secure communication on the
internet. Current TLS 1.3 connections rely on classical cryptography (RSA, ECDSA,
X25519), which are vulnerable to future quantum computers. To prepare, the
cryptographic community has standardized post-quantum algorithms like Kyber
(KEM) and Dilithium (signatures). Libraries such as OpenSSL 3 with oqs-provider and
liboqs already support PQC-enabled TLS handshakes. The challenge is to apply these
libraries to create a quantum-safe TLS termination proxy that can be implemented in
real-world environments.

Detailed Description
Your task is to implement a TLS termination proxy that uses post-quantum
cryptography for secure key exchange and authentication. The proxy should accept
incoming TLS 1.3 connections secured with PQC (Kyber, Dilithium, or hybrid
classical+PQC), then forward the decrypted traffic to the backend service (For demo
purpose local service can be used).
• Configure the TLS handshake to use Kyber KEM for key exchange and optionally
Dilithium for certificates/signatures.

• Provide a hybrid mode (classical + PQC combined) to maintain backward

compatibility.
• Ensure interoperability with standard clients (e.g., curl, browser) when using a
custom CA certificate.
• Log and compare latency and throughput benchmarks for 100 handshakes under
classical TLS vs PQC TLS.
• Supply a fallback configuration that allows classical-only TLS
Expected Solution

A complete submission will include:

• Working Proxy: Accepts TLS 1.3 PQC-enabled connections and forwards plaintext
traffic to a backend service.
• Configuration & Scripts: Easy setup with Dockerfile or one-line run command.
• Documentation: README explaining architecture, libraries used, PQC algorithms
enabled, and setup steps for testing.
• Tests & Benchmarks: Demonstrate successful PQC handshakes, failed handshake cases
(invalid certs), and performance metrics vs classical TLS.
• Security Notes: Explanation of hybrid mode, key/certificate handling, and potential
deployment considerations.

42. AI-Powered Offline Organizational Chatbot

Background:

Organisations generate and store vast amounts of Employees information across

Administration/ HR systems and employee portals. Traditional systems require
manual navigation through multiple applications. A local AI-driven chatbot can
simplify this by allowing employees, managers and HR to query data
conversationally. Further, deploying such systems requires ensuring data security,
role-based access and the solution must run offline without dependency on cloud
services.

Description:

Design and develop an offline chatbot that can answer employee queries by retrieving
information from organizational databases. The chatbot must:

• Convert organizational data into a vector database to enable efficient and accurate
question-answering capabilities.
• Implement role-based access control (RBAC) with different roles such as eemployees,
manager, administration etc
• Accuracy: The chatbot should demonstrate high factual accuracy and relevance in its
responses.

Expected Solution:

Participants are expected to deliver a fully functional offline chatbot with the following
components:

• Web Chatbot Interface

 • Data Ingestion and Vectorization: A mechanism to convert JSON data into a vector
database suitable for semantic search and question answering.
• Offline LLM Integration: A locally hosted LLM capable of processing natural
language queries and generating accurate responses based on the vector database.
• Role-Based Access Control: A secure system to authenticate users and restrict data
access based on their role (Employee, Manager or Administration/HR).
• Query Processing: A pipeline to process natural language queries, retrieve relevant
data from the vector database and generate human-readable responses.
• On-Premises Deployment: The entire solution, including the LLM and vector
database, must run locally on a single machine or a small cluster without internet
dependency.

43. Next-Gen Desktop Based Secure Data Vault

Background:
Organizations manage large volume of sensitive documents, reports, and research data.
Protecting this information with high security and confidentiality is a critical
requirement. At the same time, the organization needs to share specific data among
teams in a controlled manner. Existing cloud or browser-based vaults raise concerns
about data privacy, external dependencies, and compliance. Therefore, the organisation
requires a desktop-based secure vault solution that can run reliably within its
infrastructure. Such a system would ensure confidentiality, controlled sharing, and
compliance while empowering employees with secure, easy access to critical data.
Description:
The proposed system should function as a desktop application capable of storing all documents
in encrypted form on the local machine while also maintaining a secure copy on the server.
Any changes either made locally or on the server must be synchronized in both directions,
ensuring data consistency across environments. Since the vault will operate within the
organization intranet, it should support secure communication channels with the server while
remaining fully functional in offline mode, synchronizing data automatically when
reconnected.

To safeguard sensitive information, the system must use standard encryption algorithms and
incorporate measures to defend against ransomware attacks, such as secure backups or version-
controlled storage. It should also provide role-based and group-based access controls, allowing
documents to be shared with authorized individuals or teams. Whenever documents are shared
or updated, the concerned members should receive timely notifications or email alerts.
Additionally, the solution must support audit trails and logging, ensuring every action—such
as access, modification, or sharing—can be traced for accountability.

The ultimate goal is to deliver a secure, resilient, and easy-to-use desktop vault that protects
sensitive organizational data while enabling safe collaboration within teams.
Expected Solution:
Participants are expected to design and prototype a desktop-based secure data vault that
demonstrates the following capabilities:

• Cross-Platform Support – The application should run seamlessly on any major

operating system (Windows, Linux, macOS). Portability and compatibility are
essential for organization-wide adoption.
• Secure Local & Server Storage – All data/documents must be encrypted before
storage, both on the local machine and the server. The solution should ensure that even
if files are accessed outside the vault, they remain unreadable without proper
authorization.
• Two-Way Synchronization – Any modification made on the local machine or server
should automatically sync in both directions when connected to the intranet, while
ensuring conflict resolution and version control.
• Offline Functionality – The vault should remain fully operational even without
intranet connectivity. Users should be able to view, edit, and save documents locally,
with updates syncing once connectivity is restored.
• Strong Encryption & Security – Use of standard, proven encryption algorithms (e.g.,
AES-256, RSA) to secure documents. The system should also demonstrate basic
protection measures against ransomware-style attacks, such as immutable backups or
version history.
• Role-Based & Group-Based Access – The vault should allow administrators to create
roles and groups (e.g., directors, team heads, staff) and grant permissions accordingly.
Documents can be shared securely with groups or individuals.
• Notifications & Alerts – Whenever documents are shared, updated, or accessed, the
concerned members should be notified through in-app notifications or email alerts.
• Audit & Logs – The solution must provide logs for access, modifications, and sharing,
ensuring accountability and traceability.
• Ease of Deployment – The desktop application should be portable, lightweight, and
easy to deploy within an organization’s intranet without requiring complex setup.

Participants are free to choose the technology stack, but the final solution must demonstrate
security, usability, resilience, and cross-platform compatibility in handling sensitive
organizational documents

44. AI/ML-Driven Intelligent Software Code Analysis for Enhanced Quality and
Security
Background:
Software code analysis is crucial for detecting bugs, vulnerabilities, and inefficiencies
early in the development lifecycle. Traditional static and dynamic analysis tools often
struggle with complex codebases, evolving project contexts, and scale. AI/ML
techniques promise to improve automated code analysis by understanding code
semantics, predicting defects, and optimizing code quality. However, challenges
 remain, such as handling large project contexts, adapting to diverse coding styles,
ensuring accuracy without overwhelming developers with false positives, and
maintaining explainability and trust in AI suggestions. Especially in dynamic and
resource-constrained environments, an intelligent AI/ML solution tailored to practical
developer needs can significantly improve software quality assurance.

Detailed Description:

The challenge is to develop an AI/ML-powered software code analysis platform that can:
• Analyze large and complex codebases by breaking them into manageable units while
preserving context.
• Detect bugs, security vulnerabilities, code smells, and performance bottlenecks using
intelligent models trained on diverse coding patterns.
• Adapt to project-specific coding standards and styles automatically, providing
consistent and relevant analysis results.
• Offer explainable insights and recommendations to help developers understand and
trust the AI-based findings.
• Integrate seamlessly with existing development environments and supporting multiple
programming languages and frameworks.

Expected Solution:

An AI/ML-based smart code analysis tool that leverages advanced machine learning techniques
and natural language processing to provide automated, adaptive, and explainable code reviews.
It should include features like:
• Modular analysis for scaling to large projects.
• Real-time detection and prioritization of critical code issues.
• Customization to different development environments and languages.
• Actionable feedback for developers to fix issues efficiently.
• Integration capabilities with popular IDEs and frameworks.
The solution aims to boost software quality, mitigate security risks, reduce manual code
review efforts, and enhance overall development productivity.

45. High-Throughput Parallelized UDP Transfer with Priority Scheduling and

Adaptive Redundancy

Background
In air-gapped environments, TCP cannot be used and static Forward Error Correction (FEC)
approaches are inefficient. When transferring very large numbers of files of varying sizes,
ensuring reliable, near-lossless delivery at high throughput (1000s of files of varying sizes from
KB’s to GB’s) requires smarter scheduling and redundancy mechanisms.
Description
The proposed system should monitor a designated folder, schedule files for transfer with
priority given to smaller files over larger files (solely based on file sizes), and transmit them in
parallel over multiple UDP ports for maximum throughput. A redundancy mechanism must be
added that goes beyond static FEC, ensuring lossless transfer even under variable loss
conditions. The proposed architecture includes two machines connected back to back with one
machine designated as Transmitter (from where files will be sent) and the other machine
designated as Receiver (where files will be received) and the entire communication must
happen over UDP and should be strictly one way communication.
Expected Solution
• A working prototype that monitors a folder, schedules files based on priority, and
transmits them in parallel using UDP.
• Integrated redundancy that ensures better reliability than static FEC while maintaining
high throughput.
• Performance evaluation on large workloads (e.g., 10k files with varying sizes over 2
hours).