
Inspired by the extensibility and clean abstractions in Pytorch-lightning, this is a developer-centric backtester (most others tend to be for traders-turned-developers) that is event driven and aims to allow trading strategies to be backtested as-is relative to its forward tested counterpart.
It is highly extensible through callback hooks and supports (expects even) models for pricing/slippage, margin management, etc.
v1 completion postponed to July 2025.

Backsim: a developer-centric, event-driven backtesting framework

Pytorch-lightning inspired Backtester


A Telegram bot to track wins, losses, and statistics in your EDH/Commander Magic: The Gathering pods!
[Fully open source](https://github.com/0xEljh/edh-telegram-bot)

## Backstory
Last year my playgroup decided to try something fun: we started logging our games and wins, which started weekly goading/banter over who's been rising up the leaderboard and culminated in a little Christmas giveaway where we gave our top player a human-sized goose stuffed toy.

All this logging was done by one person who maintained a complete spreadsheet of our games. And so I thought: it would be wonderful if I could make a simple bot for this. Hence, this telegram to log games in your pod!




A Telegram bot to track wins, losses, and statistics in your EDH pods

EDH Telegram Bot


Joined Pacts, a pre-seed startup offering points-based airdrop campaigns as a service. As an early engineer, I contributed to building our unified platform for on-chain ID, activity tracking, and reward distribution. This enabled protocols to target curated, non-sybil users while providing a seamless experience for end users. My work included developing the frontend for the web app, creating a Telegram bot as an additional user interface, and implementing the reward calculation system to ensure accurate and efficient distributions based on user on-chain activity.

Pacts: A pre-seed startup helping protocols to setup and run anti-sybil points-based airdrop campaigns

Airdrops-as-a-Service


This project is an MVP LLM-based Mindful News Filter.  It helps users discover high-impact news in a focused and efficient manner.

Under the Rock News


My client wanted to test the thesis of a crypto index fund that would adequately capture the skewness of crypto returns, heavily inspired by Bassembinder.
I gathered crypto data for over 2,000+ eligible tokens spanning a decade, including many of which that would eventually go to zero (and did lots of data cleaning). 
Using a generous liquidity filter (greater than 1 million market cap), we experimented with and benchmarked various portfolio strategies, taking into reasonable account slippage, buy limits, and liquidity issues by modelling them in these backtests.
We hence overall found that buying every token at equal weight (without rebalancing) had the best overall performance (compared at various time scales, entries, etc.), short of picking the winners (BTC and ETH) directly.


Testing diversified crypto index fund strategies at size

Backtesting Crypto Index Fund


After a last-minute decision to pivot and go solo, and with less than 1 day left in the hackathon, I decided to keep it simple and play to my strengths by building a dashboard for 1inch Fusion Swaps.
This barebones dashboard achieves its primary goal of obtaining the profits made by resolvers (the executors of Fusion swap trades), the gas fees paid in the process, and the "true volume" generated by swaps.

---

This project is the final result of on-chain exploration of the Fusion Swap mechanism and resolvers. To avoid using overly dirty scrapping methods in this hackathon, we query the data via Etherscan's API. Etherscan had the data more intuitively structured for this analysis than Dune, hence this choice. Information on the ERC20 tokens and their prices was also needed, so the DeFillama historical price api was also used, especially since it supports queries via the ERC20 token address, while providing us with all the metadata like decimals, symbol, etc, hence avoiding direct eth_calls or contract calls in this poor WiFi hackathon environment.

This proof-of-concept level dashboard (just a jupyter notebook file) presents on-chain analytics of resolvers behaviour on 1inch's fusion swap, including:

- USD profit for each resolver transaction
- USD "true volume" from each resolver (no double counts on intermediate swaps)
- gas fees paid
- internal transactions made during a settlement



Analytics of 1inch Fusion Swap's resolvers

ETHTokyo 2023: #1 Dashboard


Built and backtested a series of ensemble models used to predict earnings surprises and subsequent stock price returns.

Algo-trading Earnings Calls


Slowly becoming more than just a job aggregator for crypto.

Crypto Job Board: Degen Crypto Jobs


Back when AI profile pictures were the hype, I helped drawanyone reduce the VRAM requirements for dreambooth: a few-shot finetuning method to update the model on what a "face" of our subject should look like, allowing us to then prompt images of it.
Through a mix of quantization and improving the regularization step, I halved the VRAM requirements from 24GB to 12GB, meaning we could now use 3080s instead of 3090s for training. This represented a 35% cost reduction for my client and turned them profitable.

Halved VRAM requirements for dreambooth training so it can fit 3080s

Dreambooth Training Optimization


Our goal as a startup was to become the go-to for tokenized carbon credits for the maritime industry.
For this, we first begun with facilitating the tracking of carbon emissions and performing emissions estimates based on the satellite-tracked path of vessels and the corresponding ship spec data.

Predicting emissions using satellite data and vessel specs

Maritime CO2 Emissions Modelling


At this startup focused on efficient, edge-deployed AI models (it was founded by the inventor of Extreme Learning Machines (ELMs)), I advanced from a junior engineer to leading two teams by intitiating a few key projects across our object detection pipeline.

My primary work centered around optimizing the performance of lightweight object detection models, which I did through a series of improvements: SimCLR-inspired data augmentations, loss-dependent learning rate scheduling, and a custom SGD implementation to facilitate hyperparameter tuning.
These, along with improvements to our datasets, increased our model performance on YOLOX tiny from 0.5 to 0.85 mAP across multiple datasets. I also worked on deploying these models in a quantized RKNN format (for inference on Rockchip NPUs).

The most impactful contribution I made was probably to the dataset curation process. In order to quantify the biasness of our datasets, I developed a latent space analysis and clustering, proving that the dataset had too many near-duplicates and allowing their pruning to be automated. 
Then, to make up for this loss of data and beef up our datasets, I created a semi-supervised pipeline for dataset annotation, also replacing the manual labelling exercises (which I dreaded) with a self-training system that iteratively labeled and trained a large vision model using scraped data.
Combined, these tools allowed us to rapidly curate datasets that were 3-5x larger than our original ones (and with greater diversity too!). For this, I was given two different teams to scale these tools up.

These contributions didn't just improve our R&D metrics, but also laid the groundwork for productizing our object detection solutions.

Adding tools to curate datasets and algos to break models in months made me tech lead

Low-Powered On-Edge Computer Vision


Researched the use of post-quantum encryption (FHE, MPC, etc.) in deep learning.

Internship on FHE & MPC applied to deep learning

Post-Quantum Cryptography x Deep Learning



At DSO, a defence research laboratory, I developed a Python-based finite element solver to model opto-acoustic interactions (Brillouin scattering) in high-powered laser fibres.
This tool facilitated the design of fibres with reduced power loss from backscattering and was selected for showcase from a cohort of forty.

Numerical solver to support fibre optic R&D for high power lasers

Portfolio