3+ years
Software engineering experience
Delivery across healthcare, automation products, AI workflows, and platform engineering.
Software Engineer + AI Engineer
Manoharan Thushantha Sanju
I build reliable product software and AI features that can hold up outside a demo.
My work spans React and Next.js interfaces, Node.js and Python services, delivery automation, and applied AI, with experience across healthcare software, workflow products, and ML-driven applications.
Current focus
Currently completing an MSc in Artificial Intelligence, focused on NLP, reliable prediction, and explainable machine learning.
MSc in AI
Current postgraduate study
Academic work focused on practical machine learning, NLP, and model reliability.
8 flagship builds
Selected portfolio projects
Applied AI builds and full-stack product work with clear engineering context.
Core focus
AI systems, full-stack engineering, cloud delivery
Built with React, Next.js, Node.js, Python, Docker, CI/CD, and transformer-based NLP.
About
Engineering depth across product delivery, AI systems, and reliable release work.
I do my best work where product engineering meets delivery quality. I can move from frontend UI to backend services to CI/CD and AI integration without losing focus on maintainability.
That matters when a team needs someone who can ship across the stack, make sensible technical decisions, and keep the product stable while it is still moving quickly.
Software engineering
I can move from product UI to API design to delivery workflows without losing sight of maintainability.
AI systems
My AI work stays practical: NLP pipelines, transformer models, explainability, and reliability-aware thinking.
Cloud delivery
I have worked on CI/CD, containerised delivery, monitoring, and production deployment instead of treating them as someone else’s job.
Production reliability
I care about how software behaves after release: stability, testing, observability, and real-world usability.
Flagship Projects
Work that shows how I build, not just what I used.
One project leads the story. The others show range across product engineering and applied AI.
Applied AI / Healthcare
Reliable ICD Code Recommendation
Problem
Clinical coding work is high-volume and hard to trust when a model gives predictions without context.
What I built
A transformer-based ICD recommendation workflow with uncertainty-aware output and token-level explanations for review.
Why it matters
Pushed the project beyond raw prediction by focusing on trust, interpretability, and reviewability.
Agent Infrastructure
Scout — Autonomous Research Agent
Built
A research agent built directly on Anthropic native tool use. The full agent loop is ~180 lines of Python: plan → tool-call → observe, with retry budgets, hard-fail thresholds, and three explicit exit paths. No LangGraph, no LangChain — owning the loop is the point.
Result
Tools include Tavily web search, sandboxed `python_exec` (timeout + RLIMIT, no network), URL fetch with byte cap, and per-user `remember`/`recall` memory backed by ChromaDB. Every tool call is a structlog event written to a SQLite trace store and surfaced in a per-run trace viewer. Ships with a versioned eval harness so each release has receipts, not vibes.
Full-stack AI SaaS
CVForge — Structured CV Review Powered by Claude
Built
A CV reviewer tuned for software, ML, and AI engineering roles. Full-stack SaaS with auth, billing, per-tier rate limits, revision history, and a side-by-side compare view. The technical bet: a generic AI reviewer gives bland advice; a role-aware reviewer that knows what an ML engineer’s CV should look like produces dramatically better feedback.
Result
Forces Claude into a single `submit_review` tool call with a strict JSON Schema mirrored by Zod — never parses free-text output, and re-validates at the type boundary so model surprises still get caught. Three-layer prompt architecture (base persona + per-role rubric + user context) so role rubrics can be iterated and snapshot-tested in isolation.
RAG / Retrieval
AskFastAPI — RAG Chatbot Over the FastAPI Docs
Built
A retrieval-augmented chatbot over the entire FastAPI documentation. Hybrid retrieval runs BM25 and semantic search in parallel, fuses with Reciprocal Rank Fusion, then reranks the top 20 down to 5 with a CPU cross-encoder. Markdown-header-aware chunker (~500 tokens, 50-token overlap) preserves the docs hierarchy. Streams Claude’s answer with inline `[N]` citations that link back to the exact section.
Result
25-question eval harness uses Claude-as-judge for faithfulness (structured JSON, 0–1) and parses inline citations against the retrieved set. Every release ships with retrieval@5, citation accuracy, and latency numbers — the eval harness is the headline feature, not the chatbot UI.
Systems / Devtools
Pier — Project Switcher in Rust
Built
A tiny, fast project switcher written in Rust. Replaces the morning ritual of `cd`, `source venv/bin/activate`, `set -a; source .env`, and `docker compose up -d` with a single keystroke. Single-digit-millisecond cold start. Native binary. Zero daemons. First-class support for bash, zsh, and fish.
Result
Solves the cd-from-a-child-process problem the same way zoxide does: emits a shell script on stdout that the wrapper function `eval`s in your current shell. Owns its own `.env` parser so the same file works across all three shells. 36 unit + integration tests, CI with clippy as warnings-as-errors, `thiserror` in the library and `anyhow` in the binary. v0.1.0 released.
Full-stack ML Product
Stock-Crypto Prediction Platform
Built
A MERN and Flask product that combines market views, prediction logic, and recommendation flows in one interface.
Result
Turned a model-driven idea into a usable product flow rather than a standalone experiment.
Product Engineering
Movie Ticket Booking System
Built
A full-stack booking application with JWT authentication and connected browse, select, and checkout flows.
Result
Showed full-stack ownership across auth, user flow design, and connected frontend-backend delivery.
Research-led NLP
Clinical Notes NLP Analyzer
Built
A transformer-based NLP pipeline for entity extraction and structured interpretation of clinical note data.
Result
Extended my healthcare AI work from classification into extraction and structured analysis.
Experience
Roles where I owned delivery, not just isolated tickets.
Short version: what I built, what improved, and where I had real ownership.
CueCard AI
Software Engineer
Built AI-assisted product workflows across frontend, backend, and automation.
- Built an AI sales assistant with React, Node.js, and Python.
- Cut content production time by 50% and improved engagement by 35%.
- Integrated CMS, social, and analytics APIs into a modular workflow.
Bconic
Software Engineer
Delivered workflow tooling for marketing automation and campaign operations.
- Built product features in Next.js, Node.js, and TypeScript.
- Reduced campaign setup time by 45% through workflow automation.
- Implemented email automation and real-time dashboard flows.
Level 33 Solutions
Software Engineer
Modernized healthcare software across UI, services, and platform structure.
- Helped rebuild a hospital management system using React, NestJS, and microservices.
- Improved latency by 30% through frontend and service optimization.
- Improved user satisfaction by simplifying UX and workflow performance.
OREL IT
Trainee Software Engineer
Focused on delivery automation and engineering workflow improvements.
- Built CI/CD pipelines with Docker and GitHub Actions.
- Reduced deployment time by 60%.
- Supported microservices migration work that improved reliability.
- Contributed to internal product and workflow improvements.
Technical Strengths
Capability grouped by real usage context, not arbitrary scores.
Grouped by where I have actually used the tools, not by self-scored percentages.
Languages
Used in production delivery, API work, automation, and AI projects.
Frontend
Used for product interfaces, workflow builders, dashboards, and responsive app shells.
Backend
Used for APIs, service integrations, workflow logic, and distributed application layers.
AI / ML
Used in NLP projects, transformer-based modeling, explainability work, and research builds.
DevOps / Cloud
Used to make releases faster, deployments more reliable, and systems easier to operate.
Databases / Tools
Supporting day-to-day engineering work across applications, automation, and testing.
Research & Education
Research that supports the engineering side of my work.
My academic work strengthens my engineering profile. It is focused on AI that can be explained, evaluated carefully, and used responsibly in real systems.
2025 - 2026
MSc in Artificial Intelligence and Human Factors
Coventry University
Current postgraduate work focused on NLP, reliable machine learning, uncertainty-aware prediction, and human-centred AI evaluation.
Engineering foundation
Bachelor's in Software Engineering
Sri Lanka Institute of Technology, Colombo, Sri Lanka
Built my foundation in software engineering, application development, databases, systems design, and delivery-focused engineering practice.
Current MSc focus
Transformer-based NLP and explainable AI methods that keep predictions interpretable and decision-useful.
Publication
Recommendation System for Stock and Cryptocurrency Market Using Cutting Edge Machine Learning Technology
International Research Journal of Innovations in Engineering & Technology (IRJIET) · November 28, 2023
Publication on machine-learning-based recommendation for stock and cryptocurrency market analysis.
Contact
If you need someone who can ship product work and AI delivery, let's talk.
I am open to engineering roles, applied AI work, and product teams that care about quality.