SAHIL LAB THE INTELLIGENCE LAB
Connect
BSc (Hons) Data Science Undergraduate · SUSL
SAHIL KAVISHKA

Turning Data Into Intelligence.

"Exploring AI, machine learning, language, and intelligent systems to build practical solutions to real-world problems."

Undergraduate in Data Science at Sabaragamuwa University of Sri Lanka. I treat software and machine learning as an exploratory scientific craft — designing predictive pipelines, NLP triage engines, and full-stack platforms from Anuradhapura, Sri Lanka.

Explore Projects Enter the Lab
RESEARCH TELEMETRY HUD
NODE OPERATIONAL
ACADEMIC STATUS BSc (Hons) in Data Science
Sabaragamuwa Univ. of Sri Lanka
PRIMARY DIRECTIVE Machine Learning · NLP · AI Platforms
ACTIVE BUILD SmartBuild AI (FastAPI + XGBoost + Gemini)
BASE LOCATION Anuradhapura, Sri Lanka (GMT+5:30)
PUBLIC REPOSITORIES 18
AVAILABILITY Open for Research & ML Internships
TELEMETRY TIME: —
01 // TOPOLOGY

Interactive Data Universe

Explore how core AI disciplines, mathematical modeling, and engineering systems intersect across my research and builds. Select any node to inspect telemetry and linked projects.

GRAPH v2.4 // 7 ACTIVE CONDUITS
COGNITIVE ARCHITECTURES

Artificial Intelligence

Developing agentic systems, generative intelligence, and automated decision engines that bridge mathematical models with real-world client workflows.

Verified Technologies
Gemini API PyTorch FastAPI Agentic Workflows Computer Vision
Linked Verified Systems
02 // REPOSITORIES

Experimental Project Explorer

Production platforms, research experiments, and collaborative hackathon initiatives. Inspect architecture, verified problem statements, and real codebases.

10 VERIFIED SYSTEMS
FLAGSHIP PLATFORM ACTIVE LAB BUILD

SmartBuild AI — Construction Intelligence Platform

An AI-powered architectural CAD, computer-vision floor-plan analyzer, and CIDA-indexed cost estimation platform for Sri Lanka's residential construction sector — bridging architectural studio drafting, material forecasting, and multi-bank loan feasibility in a unified system.

  • 2D/3D Architectural CAD Studio: Procedural WebGL/Three.js generation from plot dimensions and room parameters, backed by a CV floor-plan analyzer.
  • CIDA-Indexed Material BoQ: XGBoost regional price forecasting modeling non-linear supply constraints across all 25 Sri Lankan districts.
  • Banking Feasibility Engine: 45% Debt-Service-Ratio (DSR) underwriting matrix comparing BOC, Commercial Bank, HNB, Sampath, and NSB terms.
  • Agentic AI Consultant: Bilingual (Sinhala/English) civil engineering advisory agent grounded in local building standards.
View on GitHub
Python FastAPI XGBoost Three.js TailwindCSS Gemini API Docker
MICROSERVICES TOPOLOGY Client Interface (Tailwind / Three.js)
  ↓ REST Protocol
FastAPI Gateway Microservice
  ├─ 2D/3D Plan Generator Engine
  ├─ Estimator Engine (XGBoost Regressor)
  ├─ Banking Engine (45% DSR / S-Curve)
  └─ Agentic Consultant (Gemini LLM)
NLP · HACKATHON COLLABORATIVE

codeXsabra — ResoveAI

AI-powered telecom customer query triage and resolution platform engineered during the Hutch AI hackathon with NLP categorization and automated routing.

NLPFastAPIPyTorchScikit-LearnHutch AI
Source Repo ↗
CLOUD ML · DATA SCIENCE COLLABORATIVE

AWS Customer Churn Predictor

Cloud-integrated customer churn prediction engine utilizing rolling consumption velocity features, tree-based ensemble models, and AWS deployment pipelines.

AWSPythonXGBoostFlaskScikit-Learn
Source Repo ↗
FULL-STACK · PREDICTIVE ML ACTIVE

Sports Tournaments Management

Role-based web platform for organizers, players, and spectators — fixtures, live scores, leaderboards, and an ML match-outcome prediction model over REST.

DjangoMySQLScikit-learnAJAXPython
GitHub Repo ↗
NLP · TRANSFORMERS COMPLETE

Review Sentiment: ML vs. BERT

End-to-end NLP comparing classic ML against fine-tuned BERT on scraped GSMArena reviews — paired with a Chrome extension analyzing Daraz.lk reviews.

PyTorchBERTScikit-learnFastAPIChrome Ext.
GitHub Repo ↗
DATA WAREHOUSING · BI COMPLETE

Medicine Supply Chain BI

Star-schema dimensional data warehouse and Power BI dashboard tracking pharmaceutical imports, USD/LKR volatility, and hospital stockouts from 2020–2026.

Power BIPythonStar SchemaSQL
GitHub Repo ↗
ETL PIPELINE · ANALYTICS COMPLETE

EdTech Market Intelligence Pipeline

Automated data pipeline over the YouTube Data API v3 analyzing Sri Lanka's EdTech market — view velocity, channel authority, and engagement metrics feeding Power BI.

PythonYouTube APIPandasPower BI
GitHub Repo ↗
ANALYTICS · VISUALIZATION COMPLETE

Math Quiz Analytics Dashboard

Educator analytics dashboard for timed quizzes with cohort performance distributions, skill-gap detection by topic, and student leaderboards.

HTML5/JSPlotly.jsSQLPython
GitHub Repo ↗
FULL-STACK WEB COMPLETE

Salon Booking & Management

Full-stack appointment booking and management platform with administrative scheduling dashboards and client reservation workflows.

PHPMySQLJavaScript
GitHub Repo ↗
OPEN DATA · SPORTS ANALYTICS PLANNING

SL Sports Analytics Platform

An open-source Sri Lankan sports dataset initiative with statistical exploration notebooks and predictive models to unlock local sports analytics.

PythonPandasPlotlyOpen Data
Follow Repo ↗

No systems match that category or search keyword. Try clearing the filter or searching for another stack term.

03 // EXPERIMENTATION

Mini AI Laboratory

Genuinely functional, zero-latency in-browser machine learning and NLP demonstrations. No fake responses; inspect live token parsing, sentiment vectors, and 2D spatial decision frontiers.

STANDALONE IN-BROWSER EXECUTION
TEST PRESETS:
Inference: 0.00 ms (Client-Side)
LAB SPECIFICATION: Demonstrates real-time client-side lexical decomposition, negation tracking, and intent scoring in JavaScript. In production projects (e.g. Sahil's GSMArena review classifier and ResoveAI), this frontend triage connects to fine-tuned BERT and PyTorch FastAPI microservices.
SEMANTIC VECTOR READOUT
Positive Polarity (+) Confidence: 85% · Score: +0.64
Satisfaction / Joy 78%
Urgency / Severity 18%
Analytical Tone 85%
Skepticism / Risk 42%
Extracted Salient Features
#exceeds #exceptional #struggles #battery #camera
Click anywhere in coordinate space to plot new points. Feature 1 (x₁) vs Feature 2 (x₂)
KNN Hyperparameter (k) k = 3
Total Samples 12
Training Accuracy 100%
MATHEMATICAL GROUNDING: Executes client-side vector calculus in real time. The canvas renders decision partitions using k-nearest neighbors classification over a fine feature grid, displaying the spatial intuition underlying Sahil's classification models.
04 // METHODOLOGY

Inside the Model

The 7-stage machine learning lifecycle practiced across my research and engineering projects — from mathematical loss formulation to containerized deployment.

PIPELINE SPECIFICATION v3.1
STAGE 01 // OBJECTIVE MAPPING

Translating Real Constraints Into Quantifiable Loss Functions

Every AI project in the lab starts not with a model, but with defining the decision boundary and performance metrics that matter. We map business, financial, or engineering bottlenecks into mathematical optimization objectives.

REAL CASE STUDY · SMARTBUILD AI

In residential construction, material costs differ drastically across Sri Lanka’s 25 districts due to distance from Colombo supply ports. Rather than optimizing a generic national mean squared error, we partitioned regional error thresholds to prevent drastic under-bidding on cement and steel.

objective_formulation.py PYTHON 3.11 
# Define loss penalty with asymmetric under-estimation risk
def custom_construction_loss(y_true, y_pred):
    residuals = y_true - y_pred
    penalty = np.where(residuals > 0, residuals * 2.5, np.abs(residuals))
    return np.mean(penalty ** 2)
05 // CONTINUOUS EVOLUTION

Exploration Log

A transparent log of what is currently under active development, technologies I am mastering, and future horizons in physical computing and robotics.

AUTONOMOUS LAB FEED

Active Lab Builds

SmartBuild AI CAD Studio IN PROGRESS

Refining procedural 2D/3D floor-plan generation in Three.js and stress-testing computer-vision wall detection on architectural blueprints.

WebSocket Tournaments Migration IN PROGRESS

Migrating the Sports Tournaments platform from AJAX polling to bi-directional WebSockets for real-time fixture updates and match predictions.

Current Learning Vector

AWS Machine Learning Specialty STUDYING

Preparing for certification alongside cloud deployment pipelines, SageMaker feature stores, and containerized endpoints.

Advanced PyTorch & MLOps EXPANDING

Mastering model quantization, ONNX runtime export for edge devices, and automated CI/CD evaluation pipelines.

Robotics & IoT Horizon

Physical Computing Foundations EXPERIENCED

Conducted hands-on physical computing sessions with BBC Micro:bit boards for school students through IEEE WIE HOPE 2.0.

Sensor Telemetry & ROS Roadmap FUTURE HORIZON

Targeting integration of edge AI models with microcontrollers, environmental sensor matrices, and Robot Operating System (ROS) node communication.

06 // IDENTITY

About Sahil Kavishka

Data science undergraduate at Sabaragamuwa University of Sri Lanka, curious explorer, and practical problem solver.

RESEARCHER PROFILE
Sahil Kavishka — Data Science Undergraduate and AI Developer
SAHIL KAVISHKA BSc (Hons) Data Science · SUSL Anuradhapura, Sri Lanka

Bridging Computational Data & Practical Systems

I am an undergraduate reading for a BSc (Hons) in Data Science at Sabaragamuwa University of Sri Lanka. Rather than confining myself to a single language or framework, I view technology as a connected ecosystem. My work sits at the intersection where analytical machine learning models meet robust full-stack engineering.

Whether structuring dimensional data warehouses to model national medicine stockouts during economic shocks, training gradient-boosted regressors for regional construction estimates, or packaging NLP sentiment classifiers into live browser extensions, I focus on building systems that solve tangible problems.

I value intellectual curiosity, experimentation, clean architecture, and continuous learning above all else.

18
Public Repos
SUSL
Data Science
GMT+5:30
Timezone
<24h
Response Time
Core Philosophy: "Without data, you're just another person with an opinion." — W. E. Deming.
EVOLUTION // FULL-STACK DEV → DATA SCIENTIST

Engineering Journey & Growth Trajectory

How practical web foundations evolved into mathematical data science, predictive pipelines, and agentic AI systems.

STAGE 01 // FOUNDATIONS

Full-Stack Web Foundations

Started with practical, client-facing web applications. A PHP & MySQL salon booking system taught core CRUD operations, database indexing, authentication, and administrative dashboards.

PHPMySQLJavaScriptAuth / CRUD
STAGE 02 // SCALE

Role-Based Platforms at Scale

Transitioned to Django and React to build the Sports Tournaments platform — architecting multi-role access controls, automated tournament brackets, and live match scoring for organizers, players, and spectators.

DjangoPythonReactRole-Based Systems
STAGE 03 // DATA SYSTEMS

Layering Machine Learning & BI

Layered machine learning and dimensional warehousing onto operational software — sports match-outcome classification, quiz cohort analysis, and a Star-Schema Power BI warehouse tracking Sri Lanka's medicine supply chain.

Scikit-learnPower BIStar SchemaPandas
STAGE 04 // ACTIVE NOW

AI-Agentic & Multimodal Platforms

Synthesizing LLM agents, computer vision, and gradient-boosted regressors into SmartBuild AI — while mastering PyTorch, containerized MLOps, WebSocket concurrency, and edge physical computing.

FastAPIGemini APIXGBoostThree.jsMLOps
CAPABILITIES // GROUPED BY PURPOSE

Technical Competencies

Core Languages
Python SQL JavaScript (ES6+) R Bash PHP
Machine Learning & AI
Scikit-learn XGBoost PyTorch Hugging Face (BERT) Gemini API NLTK
Data Analytics & Pipelines
Pandas NumPy Power BI Star Schema PostgreSQL YouTube API
Full-Stack & Systems
FastAPI Django Three.js Docker Git / GitHub Micro:bit
TOOLKIT // DAILY OPERATIONAL ENVIRONMENT

Active Toolkit & Workstation Instruments

The actual software, command-line environments, and instruments open on my machine while engineering.

EDITOR PRIMARY

VS Code

Configured with Pylance, ESLint, and GitLens for strict type checking and commit provenance.

VERSIONING CI/CD

Git + GitHub

Disciplined feature branching, semantic commits, and PR review habits across all repositories.

API TESTING REST SUITE

Postman

Dedicated integration collections and automated schema assertions for FastAPI and Flask endpoints.

CONTAINERS PARITY

Docker

Local development parity and reproducible environments for multi-service Python microservices.

NOTEBOOKS ANALYSIS

Jupyter

Rapid mathematical experimentation, feature correlation, and hyperparameter sweeps before API export.

BI & WAREHOUSE STAR SCHEMA

Power BI

Interactive data visualizations, star schema relationships, and executive reports for stakeholders.

DESIGN PROTOTYPING

Figma

Information architecture diagrams and UI wireframes before writing frontend stylesheets and HTML.

SHELL LINUX-FIRST

WSL2 + zsh

Linux-first shell environment with customized scripts for data munging, SSH, and server management.

GITHUB // TELEMETRY & ACTIVITY

Live Code Activity & Open Source Telemetry

REPOSITORY TELEMETRY ● ACTIVE DEV
Public Repositories 18
Development Focus Machine Learning · Data Systems
Code Review Workflow Feature Branches · Semantic PRs
Primary Profile @sahilkavishka ↗
PRIMARY LANGUAGE DISTRIBUTION CODEBASE METRICS
Python 48%
JavaScript / ES6+ 24%
SQL / Power BI 16%
PHP / HTML / CSS 12%
YEARLY COMMIT MATRIX // CONTRIBUTION CALENDAR Inspect Profile on GitHub ↗
Sahil Kavishka's GitHub Contribution Calendar
LEADERSHIP // ACADEMIC & COMMUNITY HONORS

Leadership & Competitive Finalist Credentials

EXECUTIVE LEADERSHIP 2025 – Present

Secretary — Society of Computer Sciences

Sabaragamuwa University of Sri Lanka

Leading executive student governance, faculty communications, technical workshops, and industry networking initiatives for the university computing and data science student community.

HONOR & AWARD AWS Community

Certificate of Outstanding Achievement

AWS Student Builder Group · SUSL

Awarded for active engagement, valuable contributions, and community leadership in promoting AWS cloud computing and tech innovation among undergraduates.

STEM OUTREACH HOPE 2.0

STEM & Micro:bit Coding Trainer

IEEE WIE · Karagasthalawa School

Conducted hands-on physical computing sessions with BBC Micro:bit boards for school students, introducing sensor telemetry and algorithmic thinking.

ORGANIZING TEAM NEXORA 1.0

Program Team Member — NEXORA 1.0

IEEE Student Branch · IEEE CS SUSL

Contributed to program conceptualization, session logistics, and event coordination for NEXORA 1.0, a premier technology leadership conference.

MENTORSHIP Code Club

Volunteer Coding Mentor

Code Club Sri Lanka · Ratnapura Library

Mentored young learners in foundational programming, visual problem-solving, and digital literacy to empower youth through accessible tech education.

COMPETITIVE FINALS Team Code Wave

Competitive Programming Finalist

J'Pura Xtreme 3.0 & CodeRally 7.0

Qualified as University Category Finalist in J'Pura Xtreme 3.0 (IEEE USJ / Buildr Labs) and 24-Hour Grand Finale Finalist in CodeRally 7.0 (IEEE CS SUSL).

07 // FIELD NOTES

Research Dispatches & Engineering Notes

Documenting decisions, architectural trade-offs, and empirical findings — focusing on why systems are built the way they are.

TECHNICAL PREPRINTS
AGENTIC AI · ARCHITECTURE FIELD NOTE 01

Building SmartBuild AI: What "Agentic" Actually Means in a Construction Estimator

Why the banking feasibility underwriting matrix and the conversational AI consultant operate as isolated microservices rather than a single monolithic prompt — and what broke when I first tried it the other way.

Focus: Multi-agent boundaries · FastAPI
NLP · BENCHMARKS FIELD NOTE 02

Classic ML vs. Fine-Tuned BERT: Where Complexity Earns Its Latency

A practical benchmark on scraped GSMArena smartphone reviews — evaluating inference latency against F1-score gains to find where a Random Forest is enough and where transformers are non-negotiable.

Focus: PyTorch · Latency vs Accuracy
BI · DIMENSIONAL WAREHOUSING FIELD NOTE 03

Designing a Star Schema to Survive Hospital Emergency Decision-Making

Lessons from modeling Sri Lanka's pharmaceutical import manifests, foreign exchange volatility, and district medical supply distributions into a Power BI model that withstands high-pressure queries during national stockout risks.

Focus: Star Schema · Real-world constraints
Interested in technical write-ups and benchmark releases?

Receive an email dispatch when new architectural analyses or benchmark preprints are published.

Get Notified by Email
08 // TRANSMISSION

Initialize Connection

Open for research collaborations, Data Science & Machine Learning internships, and technical engineering opportunities.

DIRECT DISPATCH CHANNELS

Direct Channels

Email is the fastest medium to reach me. You can also connect via LinkedIn or inspect ongoing code updates on GitHub.

Direct Email sahilkavishka428@gmail.com GitHub Profile @sahilkavishka LinkedIn Profile in/sahilkavishka Kaggle Notebooks @sahilkavishkakumara

Transmit Message

Generates a pre-filled transmission draft in your preferred mail client.