sh Shalom Hunukumbura

Available for conversations · Malabe, Sri Lanka

Shalom Hunukumbura
builds the boring parts that have to be right.

Full stack developer shipping taxation, accounting and compliance platforms end to end: schemas, REST APIs, frontends and releases. Outside work I build AI tools for finance and publish open-source packages, four of them on PyPI.

15production services contributed to
140+REST endpoints designed and integrated
240+merged pull requests
46Alembic migrations authored

01

Selected work

Things I built on my own time, each solving a problem I ran into at work or in Sri Lanka.

  1. lk-tax

    Sri Lankan tax engine, on PyPI

    APIT, income tax, withholding tax, VAT and SSCL, with rules stored as dated data that cites its Inland Revenue source, so any date since 2023 uses the rules in force then. Verified against all 75,590 rows of the official salary tax tables, and runs fully in the browser as a live calculator.

    • Python
    • Decimal
    • FastAPI
    • Pyodide
    • PyPI
  2. lk-tax-mcp

    AI tool server, in the official MCP Registry

    An MCP server that gives AI assistants like Claude 9 exact, sourced Sri Lankan tax tools instead of recalled rates. Designed for an AI as the user: described parameters, enumerated choices, and errors the model can act on.

    • Python
    • MCP
    • LLM tools
    • PyPI
  3. payhere-sim

    PayHere payments testing, on PyPI

    Sends correctly signed PayHere payment notifications to localhost and checks a webhook handler against 10 scenarios: forged, underpaid, duplicate, late and simultaneous. Caught a race that granted 500 credits for one payment in a handler that passed manual testing.

    • Python
    • Webhooks
    • Concurrency
    • PyPI
  4. alembic-guard

    Migration linter for PostgreSQL, on PyPI

    Catches Alembic migrations that lock Postgres tables or break running code before they reach production. Parses migrations with Python's ast (never executes them), ships 10 rules with suggested fixes, and runs as a CLI or a GitHub Action that annotates pull requests.

    • Python
    • Alembic
    • PostgreSQL
    • AST
    • GitHub Actions
    • PyPI
  5. DocSage

    RAG from first principles, no LangChain

    Question answering over 37 engineering documents with cited answers. Hybrid retrieval (FAISS embeddings + BM25, merged with reciprocal rank fusion) puts the right document first for 93% of questions vs 86% for either alone, measured on a 54-question eval set. Refuses off-topic questions instead of guessing.

    • Python
    • FAISS
    • BM25
    • FastAPI
    • LLM
    • Docker

Earlier: Actify (meeting notes to action items with Gemini), Customer Churn Prediction (ML pipeline + FastAPI) and more on GitHub.

02

Experience

One company, two platforms, from intern to owning features across 12 services.

Aug 2025 – Present

Xaventra

Software Engineer, Full Stack Developer

BizAdvisor platform

  • Build and ship features across a 15-service production platform covering taxation, accounting, incorporation and compliance, authoring 170+ merged PRs on a 6-engineer team.
  • Design and integrate REST APIs across FastAPI and Flask services on PostgreSQL, contributing to 140+ endpoints consumed by Next.js / TypeScript frontends.
  • Own features end to end across 12 services, including invoicing, payments, promotions and incorporation workflows, from schema and business logic to UI and release.
  • Diagnose production incidents through log analysis, database inspection and API tracing, restoring service with minimal downtime and documenting root causes.
  • Author 46 Alembic migrations and maintain 8 GitHub Actions CI/CD workflows, including automated migration checks for safe, repeatable deployments.
  • FastAPI
  • Flask
  • PostgreSQL
  • Next.js
  • TypeScript
  • Alembic
  • GitHub Actions

Feb 2025 – Aug 2025

Xaventra

Software Engineering Intern, Full Stack Developer

BizAdvisor and Neurux platforms

BizAdvisor, a compliance platform for corporate secretarial firms

  • Developed backend services with Flask and SQLAlchemy, integrated with React / TypeScript frontends for multi-party task coordination workflows.
  • Fixed intricate ORM relationship and data-binding defects, correcting coordinator attribution and inaccurate UI rendering.
  • Refactored legacy components and resolved production issues in an agile team.

Neurux, a domain-focused AI assistant for business and compliance

  • Adapted an AnythingLLM fork into a domain-specific chatbot for finance and compliance users.
  • Implemented RAG pipelines with embeddings and structured document retrieval, improving contextual relevance and precision.
  • Iterated on prompt structures to reduce off-domain answers, and prototyped agentic workflows for multi-step tasks.
  • Flask
  • SQLAlchemy
  • React
  • RAG
  • Embeddings
  • Prompt engineering

03

Skills

What I reach for, grouped by layer.

Languages
Python, TypeScript, JavaScript, SQL, HTML, CSS
Backend
FastAPI, Flask, SQLAlchemy, Node.js, Express, REST API design
Frontend
React, Next.js, TypeScript, Tailwind CSS, responsive UI
Databases
PostgreSQL, MySQL, MongoDB, SQLite, schema design and migrations
AI and ML
RAG, LLM integration, prompt engineering, embeddings and vector retrieval, MCP servers, agentic workflows, scikit-learn
Cloud and tools
AWS (S3), Linux, CI/CD, GitHub Actions, Vercel, Docker, PyPI packaging, Git, Postman, JWT auth, AI-assisted dev tooling

04

Education

  • BSc (Hons) in Information Technology

    Cloud Computing · Sri Lanka Technological Campus · 2025

  • AWS Cloud Practitioner training

    Completed

05

Learning log

A monthly record of the books and articles I read, what I learned, and how I applied it at work and in side projects.

Read the log

Have a platform that needs to be right? Let's talk.

Open to full stack and AI engineering conversations.

benjaminshalom1999@gmail.com