Software Engineer

Janith Chathusha

I build full-stack products and practical AI systems.

Focused on engineering craft — scalable backend architecture, explainable machine learning pipelines, and intelligent automation that earns its place in production.

Explore Projects Download CV
Based in
Sri Lanka
Now
Absolx
Focus
Full-Stack · AI
Portrait of Janith Chathusha

01 Selected Projects

Systems that turn noisy signals into clear decisions.

Six specialized engineering builds spanning deep learning explainability, computer vision kinematics, sports predictive modeling, and embedded IoT sensing — presented in glassmorphic telemetry cards.

newsshield.app/analyzeARTICLE · TOKEN ATTRIBUTIONFAKEREALBREAKING:Scientistsconfirmmiraclecurehiddenbyofficials,sourcessaytheshockingtruthwassuppressedforyearswhileindependentexpertsremainsilentasviralclaimskeepspreadingonlinetoday.TokenizeEmbedBiLSTMSoftmaxLIME▲ Sensational framing + unverified sourcing drive the prediction.VERDICT87%LIKELY FALSETOP ATTRIBUTIONSBREAKING+0.95miracle+0.90shocking+0.85suppressed+0.75experts-0.50independent-0.45
Token attribution heatmap · LIME explainability FIG. 01

01 NLP · Full-Stack AI

NewsShield

Explainable Fake News Detection

Built a web application that evaluates news content using a BiLSTM model, rule-based checks, and explainable AI. Instead of showing only a classification verdict, NewsShield highlights the exact words that influenced the result, helping users inspect and understand the model’s reasoning.

Key Highlights

  • ◆ News classification with continuous prediction confidence scoring
  • ◆ Word-level explainability using LIME (Local Interpretable Model-agnostic Explanations)
  • ◆ Rule-based heuristics for identifying misleading framing and suspicious sourcing

Technologies

ReactFramer MotionFastAPITensorFlowKerasLIME
LinkedIn Case Study Inquire Architecture
LPL 2026 · MATCH 14 · LIVEmodel v3.2 · 240msBATTING142/4(16.0 ov)WIN PROBABILITY71%▲ 5.20255075100OVER 120KEY DRIVERSRun-rate delta0.82Wickets in hand0.64Venue par score0.48Bowler economy0.36SIMULATION10k Monte Carlo runs
Win-probability telemetry · Gradient × Input attribution FIG. 02

02 Machine Learning · Sports Analytics

LPL 2026 Intelligence

Explainable Cricket Analytics Platform

Developed an interactive cricket analytics platform that combines live match predictions, player performance tracking, and tournament simulations. Connects machine learning models with an explainability layer so analysts and coaches can explore both the predictions and the factors behind them.

Key Highlights

  • ◆ Real-time match win-probability forecasting and projected innings scores
  • ◆ Player performance indexing across match phases and head-to-head dynamics
  • ◆ Tournament forecasting powered by 5,000 Monte Carlo match simulations
  • ◆ Feature contributions visualized using Gradient × Input explainability

Technologies

PythonPyTorchscikit-learnPandasNumPyFastAPIStreamlitMatplotlib
LinkedIn Case Study Inquire Architecture
MEDIAPIPE POSE · FRAME #148 · RELEASE DETECTED30 FPS · 33 LANDMARKS168.4°ESTIMATED POSE · 2D LANDMARK PROJECTION● RELEASE POINT DETECTED AT T = 1.24s (FRAME 37)KINEMATICS TELEMETRYBowling Arm Elbow Angle168.4°+2.1° toleranceShoulder Counter-Rotation28.2°front-on alignmentFront Knee Flexion Angle154.0°brace-impact phaseTrunk Lateral Flexion14.8°release alignmentRELEASE VELOCITY VECTOREstimated Release Speed134.6 km/h (est.)* Experimental video estimation · Not calibrated radar
Markerless pose tracking · Joint kinematics overlay FIG. 03

03 Computer Vision · Video Analytics

Experimental Prototype

Cricket Biomechanics

Bowling Action Analyzer

Created a computer vision prototype that analyzes cricket bowling video footage using markerless pose estimation. It tracks body landmarks and joint kinematics, identifies ball release frames, and produces annotated video alongside graphical reports for technique exploration.

Key Highlights

  • ◆ Markerless pose tracking and real-time joint-angle visualization
  • ◆ Automated bowling action phase detection and ball release-frame capture
  • ◆ Kinematic trajectory and release angle metrics for biomechanical review
  • ◆ Annotated video overlays and frame-by-frame joint angle progression charts

Technologies

PythonOpenCVMediaPipe PoseNumPyPandasMatplotlib
Inquire Architecture
MATCHMIND · ELO + POISSON MODEL · WORLD CUP SIMULATION10,000 ITERATIONSSPAINELO: 2042 · FORM: W-W-D-WxG 2.14VSGERMANYELO: 1988 · FORM: W-D-W-LxG 1.48OUTCOME PROBABILITY DISTRIBUTION48.2%27.6% DRAW24.2%● PREDICTED WINNER: SPAIN (+0.66 xG DELTA)KNOCKOUT BRACKET SIMULATIONESP vs SUIESP (3 - 1)GER vs DENGER (2 - 0)ESP vs GERESP (2 - 1)SEMIFINALESP proj.MOST LIKELY SCORELINES2 - 114.8%1 - 113.2%1 - 011.5%2 - 09.4%MODEL METHODOLOGY● Elo Rating Differential● Poisson Goal Distribution● Expected Goals (xG) Momentum● Historical H2H Weighting50,000 Monte Carlo sims
Match forecast engine · Tournament bracket progression FIG. 04

04 Predictive Modeling · Web Application

MatchMind

Football Prediction & Tournament Explorer

Built a football match prediction platform that transforms team statistics, Elo ratings, recent form, and historical head-to-head performance into interactive match forecasts and tournament progression projections.

Key Highlights

  • ◆ Head-to-head team comparisons with win, draw, and loss outcome distributions
  • ◆ Poisson-based expected goals (xG) projections and most likely scorelines
  • ◆ Interactive knockout bracket simulator and team progression insights
  • ◆ Methodology transparency layer explaining feature weights and ratings

Technologies

PythonPandasscikit-learnjoblibHTML5CSS3JavaScript
LinkedIn Case Study Inquire Architecture
EMBEDDED HARDWARE PROTOTYPE · SAMPLING 1Hz · ACADEMIC RESEARCHBAUD: 960016x2 ALPHANUMERIC LCD INTERFACEFRUIT: BANANA D4T:28C H:78%MQ4:142 MQ3:086ppmTIME-SERIES SPOILAGE DECAY CURVE (72 HOURS)SPOILAGE THRESHOLD (100 ppm)MQ-4 (Methane)MQ-3 (Alcohol)0h (FRESH)36h (RIPE)72h (OVERRIPE)SENSOR CHANNELSMQ-4 (Methane / CH4)142 ppmElevated fermentationMQ-3 (Alcohol / C2H5OH)86 ppmEarly spoilage thresholdDHT11 Temperature28.4 °CAmbient storageDHT11 Humidity78.2 %High moisture zoneHARDWARE ARCHITECTURE● Microcontroller MCU (ADC)● MQ-3/4 Heated Analog Sensors● 16x2 HD44780 Liquid Crystal
Hardware sensor telemetry · VOC concentration curves FIG. 05

05 IoT · Embedded Systems · Research

Academic Research Prototype

Food Quality Checker

Sensor-Based Freshness Monitoring

Contributed to an academic food-quality monitoring prototype that uses gas sensors and environmental telemetry to detect chemical markers associated with fruit freshness and spoilage over time.

Key Highlights

  • ◆ Multi-channel gas telemetry measuring volatile organic compounds (MQ-3, MQ-4)
  • ◆ Environmental monitoring capturing ambient temperature and relative humidity (DHT11)
  • ◆ Microcontroller firmware with button-controlled mode selection and 16x2 LCD readout
  • ◆ Experimental time-series data logging for freshness degradation analysis

Technologies

Embedded C/C++MQ-3MQ-4DHT11MicrocontrollerHardware Interfacing
Inquire Architecture
N8N ORCHESTRATION PIPELINE · SUPABASE POSTGRESQL · ACTIVEAVG LATENCY: 280msAUTOMATION WORKFLOW DAG (N8N)WebhookTriggerAuth & SanitizeSecuritySupabase QueryPostgreSQL DBMemory BufferSession CacheLLM OrchestratorReasoning EngineEXECUTION TELEMETRY & MEMORY LOG12:04:18Incoming user query: "Order #8421 tracking & refund policy"12:04:18Supabase: SQL query returned 1 matching order row (12ms)12:04:19Memory: Retrieved 4 prior conversational turns from session12:04:19LLM Response generated with grounded policy citations (240ms)PIPELINE METRICSGrounding Precision99.2%Zero hallucination on DB schemaMedian Query Latency280msSub-second customer response timeSTACK ARCHITECTURE● n8n Node Workflow● Supabase (PostgreSQL)● Conversational Memory● Dynamic Prompt Retrieval● Webhook Event Emitters● Error Fallback Queue● REST API Endpoints
n8n workflow orchestration · Database integration pipeline FIG. 06

06 AI Automation · Database Integration

External Work

AI Customer Assistant

Enterprise Workflow Automation

Contributed to production-tested AI customer-assistant workflows connecting natural-language conversational interfaces with relational business data, vector search, and automated back-office execution tools.

Key Highlights

  • ◆ Autonomous chat workflows wired to PostgreSQL databases via Supabase
  • ◆ Dynamic prompt templates, conversation memory buffers, and context retrieval
  • ◆ Rigorous evaluation benchmarks on product catalog, pricing, and availability queries
  • ◆ Webhook event listeners and automated exception-handling queues

Technologies

n8nSupabasePostgreSQLLLM APIsWebhooksREST APIs
Inquire Architecture

02 External & Enterprise Work

Production engineering, described by pattern.

Client and enterprise production systems are strictly anonymized. What remains is the architecture — resilient patterns designed for high throughput, sub-second latency, and fault tolerance.

  1. 01 /

    Enterprise

    Autonomous AI Agent Platform

    Architected multi-agent execution queues and autonomous workflow pipelines connecting LLMs with external tools, stateful memory buffers, and asynchronous task telemetry.

    Architecture

    FastAPI · PostgreSQL · Redis Queue · Worker Nodes · WebSocket

    Agent QueuesTool CallingAsynchronous OpsLangGraph

  2. 02 /

    Enterprise

    Distributed Commerce & Fulfillment Ecosystem

    Engineered high-throughput catalog indexing, real-time inventory synchronization, and resilient checkout pipelines designed for sub-second page loads and zero inventory collision.

    Architecture

    Next.js · Node.js · InfluxDB / PostgreSQL · Redis · Docker

    High-Throughput SSRInventory SyncDistributed CartMicroservices

  3. 03 /

    Enterprise

    Executive Media & Editorial Content Engine

    Built ultra-fast SSR content delivery infrastructure with headless CMS integration, structured SEO graph generation, and asset optimization pipelines for national business leadership media.

    Architecture

    SvelteKit / Next.js · Headless CMS · CDN Edge · AWS S3

    Headless CMSEdge SSRAutomated SEOPerformance

03 Living Architecture

The Neural Skill Tree.

A living computational organism: foundational logic roots, high-throughput systems trunk, and four specialized intelligent branches rendered in real-time volumetric depth.

NEURAL_TREE // SYNAPTIC CONDUCTION
The Neural Tree of Engineering
Roots // Foundational Logic:

AI & Agentic Systems

Active Cluster

Multi-agent autonomous systems, Explainable AI, and neural architectures.

Synaptic Sub-Systems (Select to Inspect):

PRODUCTION_CONTEXT // TELEMETRY
Specialized LEVEL

Agentic AI & Orchestration

VERIFIED NODE

Autonomous collaborative multi-agent decision systems engineered for client workflows and workflow synthesis.

Core Frameworks & Toolchain:

LangGraph CrewAI LangChain n8n

Status

Production

Architecture

Distributed

Integration

Active

04 Experience

A steady line from trainee to engineer.

  1. Mar 2026 — Present

    Software Engineer (Full-Stack)

    Absolx

  2. Nov 2025 — Feb 2026

    Associate Software Engineer (Full-Stack)

    Absolx

  3. Apr 2025 — Dec 2025

    Associate Data Analyst

    Inplay Matrix

  4. Feb 2024 — May 2024

    Software Engineer (Contract)

    Nextzela Technologies

  5. Feb 2023 — Aug 2023

    Trainee Software Developer

    Bank of Ceylon

05 About

Engineering is the discipline of making signal useful.

I work across the full stack — from data pipelines and model serving to the interfaces people actually touch. The thread through all of it: build things that hold up in production and explain themselves.

Janith Chathusha — SLIIT Convocation
SLIIT Graduate · 2024

BSc (Hons) in Information Technology

SLIIT Convocation · Sri Lanka

Academic Foundation

BSc (Hons) in Information Technology

Sri Lanka Institute of Information Technology (SLIIT)

2020 — 2024

Graduated with a strong foundation in distributed systems, algorithm design, software architecture, and artificial intelligence. Guided by a constant drive to transform theoretical computing into deployed software that serves real human needs.

Craft
Readable code, explicit contracts, and systems that are pleasant to change.
Scale
Backends designed around failure modes, queues, and observability from day one.
Usefulness
AI only where it measurably helps — explainable, testable, and owned.

06 Initiate Transmission

Let's engineer something extraordinary.

Specialized in scalable backends, explainable machine learning systems, and high-performance product architectures. Whether you're architecting a new platform, seeking applied AI expertise, or assembling an ambitious engineering team, my transmission channel is open.

Transmit Direct Email Download Curriculum Vitae
Available for Roles & Teams
Q4 2026
Base Station
Colombo, Sri Lanka
Local Time (UTC+5:30)
Response SLA
< 24 Hours Guaranteed
Primary Focus
Full-Stack · Practical AI
Janith Chathusha ONLINE

Associate Software Engineer @ Absol X · Full-Stack & AI

Janith Chathusha · Just now

Hello! I am Janith Chathusha.

Welcome to my portfolio. I specialize in Full-Stack Engineering, Agentic AI Architectures (LangGraph, CrewAI), and Scalable Distributed Systems. Whether you have questions regarding my production experience, systems architecture, my research project NewsShield XAI, or custom engineering solutions — feel free to ask me anything in English or සිංහල.

How can I help you today?

Prompts:
Direct Client Communication

Initiate Instant Connection.

Reach out directly for engineering roles, technical consultations, or agentic AI architectures via WhatsApp or direct transmission.

WhatsApp Direct

+94 783356601

Instant client communication · Chat now ↗
Direct Email Channel

chathushadayawansha1346@gmail.com

< 24h SLA · Encrypted transmission

Open Mail App

Direct Transmission Protocol

Dispatches straight to Janith Chathusha's primary inbox.

Active Channel

Direct: chathushadayawansha1346@gmail.com