Software Developer & AI/ML Engineer
I build software
that ships.
I'm Alston Mendonca. I turn machine-learning models and full-stack interfaces into production software real users depend on every day.
Selected work
RConnect
Quant options backtesting terminal · NSE derivativesA QuantConnect/LEAN-style event-driven options backtesting engine for Indian index derivatives (NIFTY, BANKNIFTY, SENSEX), wrapped in an Electron terminal with paper trading and a live data pipeline. ~48k LOC of Python/FastAPI with margin, Greeks, optimizer, and walk-forward, hardened by 186+ audited quant-correctness fixes, a 796-test suite, and fill friction calibrated against 196.9M LSEG quotes over a 94M-row DuckDB store.
ViperCore
Offline-first restaurant POS · in productionCo-founded and built a fully-fledged offline-first Point-of-Sale platform running in production at cafes in Karnataka, India. Architected the desktop app with Next.js and Electron.js, featuring AES-256 encrypted local SQLite storage, 9 production modules (billing, KOT printing, inventory, tax reporting, employee analytics), and GST-compliant invoicing.
DueLLM
Self-refining LLM orchestrationA self-refining generative-AI system where two LLMs iteratively critique and improve each other's outputs, with multi-model orchestration via Amazon Bedrock. Full-stack app built with Next.js and FastAPI with real-time streaming, driving automated quality improvement across reasoning, code generation, and creative writing.
DressGuard AI
Real-time clothing compliance · YOLOv8Led a team of 4 to build a real-time AI-powered clothing detection and compliance system supporting webcam, video, image, and IP-camera inputs. Fine-tuned YOLOv8 on custom datasets, built a configurable rule engine with automatic model discovery and bounding-box visualization, and added facial detection and recognition via FaceNET and Redis.
ProperLCP
Restaurant POS · in productionLed a team of 2 to design, develop, and deploy a full restaurant Point-of-Sale system from scratch: 75+ core features including real-time billing, inventory tracking, and reporting dashboards comparable to PetPooja and Dineout. In production use at The Lassi Corner, SJEC, Mangaluru.
Archive
A cross-platform mobile app supporting a year-long PhD study on mental wellness among nurses, with structured surveys, emotional check-ins, journaling, and guided mindfulness audio on a scalable Supabase backend.
A voice-driven data-visualization tool that converts spoken natural-language queries into interactive charts and graphs using generative AI.
I build things that work in the real world.
Full-stack foundation, machine-learning specialism, a bias toward shipping.
I'm a Bengaluru-based Software Developer & AI/ML Engineer with 1.5+ years of professional experience building and deploying machine-learning systems. I specialize in NLP, Computer Vision, and LLM integration, from React/Next.js frontends to Node.js and FastAPI backends.
I co-founded ViperCore, an offline-first POS platform running in production; fine-tuned BERT to near-99% accuracy in production on AWS SageMaker; and shipped compliance systems on YOLOv8. I like turning complex ideas into production software people rely on every day.
Where I've worked
Junior Quantitative Developer - Indian Options & Backtesting Systems
RPG Trading DMCC
- Built RConnect, an event-driven options backtesting engine (QuantConnect/LEAN-style) for NSE index derivatives: ~48k LOC of Python/FastAPI with margin, Greeks, optimizer, and walk-forward, backed by 796 tests.
- Ran adversarial quant audits and fixed 186+ correctness defects: look-ahead bias, intrabar stops, end-of-day mark-to-market loss, and margin double-counting, each locked in with regression tests.
- Calibrated fill friction against 196.9M LSEG bid/ask quotes into a moneyness- and VIX-scaled per-leg model, and re-fit brokerage to the firm's real cost structure.
- Designed a GIFT-vs-NSE convergence arbitrage backtest on synthetic put-call-parity futures, behind a 303GB LSEG tick-to-parquet pipeline and a 94M-row DuckDB store.
- Shipped the terminal as an Electron app (React/TypeScript, lightweight-charts, AG Grid, Monaco) with an AI sidebar assistant on the Claude Agent SDK.
Co-Founder & Chief Technology Officer
- Co-founded ViperCore, an offline-first Point-of-Sale platform running in production at cafes in Karnataka, covering real-time billing, inventory management, GST-compliant invoicing, and analytics dashboards.
- Architected and migrated the full-stack desktop app from Electron.js + HTML to Next.js with an Electron.js backend, using an offline-first architecture with AES-256 encrypted local SQLite for 100% offline reliability.
- Designed and deployed 9 core production modules: real-time billing, table management, KOT printing, a visual receipt editor, inventory tracking, tax reporting, employee analytics, and data export.
- Led the end-to-end product lifecycle from concept to production, managing on-site installations and direct support for restaurant clients.
Backend Developer & Machine Learning Engineer Intern
Aquera
- Developed and fine-tuned a BERT-based transformer for automated service-account classification within Identity Access Management, reaching near-99% accuracy and deploying to production via AWS SageMaker.
- Optimized ML pipelines for attribute-mapping analysis, cutting end-to-end processing time by 10% and improving classification accuracy by 30% through better feature engineering and preprocessing.
- Built and evaluated LightGBM, CatBoost, RuleFit, and Elkan-Noto PU-Learning classifiers for identity resolution and access governance.
- Integrated LLM capabilities into existing data-processing workflows on AWS for intelligent document parsing and automated query resolution.
Software Developer & AI Intern
Heritage Safety Training Centre
- Designed and built the company's responsive website with React.js and Vite, improving experience and accessibility for training-course participants.
- Deployed and integrated a Moodle-based LMS into the website, supporting 500+ registered users for course delivery, enrollment, and training-workflow automation.
- Built an LLM-powered conversational chatbot using ChatGPT with custom system prompts and retrieval-augmented generation for automated user support.
The toolkit
Languages
Frontend
Backend
Databases
AI / ML
Cloud & DevOps
Where I studied
St. Joseph Engineering College
B.E. Artificial Intelligence & Machine Learning
Relevant courseworkDeep Learning · Mathematics for ML · Data Structures & Algorithms · DBMS · Computer Vision · Statistics & Probability
GEMS Our Own Indian School
Senior Secondary (Class XII), CBSE Board
Résumé
Let's talk
Open to AI/ML and full-stack roles, collaborations, and interesting problems. The fastest way to reach me is email, and I reply within a day.