Backend engineering / independent research

Sharjeel Abbas

I build backend systems and study the engineering decisions behind them. My current work focuses on data-local AI memory, retrieval, and code generation.

Lahore, Pakistan · Get in touch ↗

Selected work

All work →
01

TelloDB

A data-local memory engine for AI agents. One Rust process combines temporal fact versions, lexical and vector retrieval, graph relations, and source-aware answers.

Visit TelloDB ↗
02

RigorDesk

A multi-LLM peer-review platform with seven debate protocols, stored rounds, pairwise scoring, and replayable verdicts. It integrates more than 100 models across 20 providers.

Visit RigorDesk ↗
01

TelloDB

A coverage-aware multi-index memory engine for data-local AI agents

Read paper ↗
02

Compact recurrence encodings for code generation

A four-model study of program length, correctness and token usage

Read paper ↗
October 2026

Shorter code can be harder to generate

A compact notation saves characters on the page. That does not guarantee fewer tokens spent reaching a correct program.

Read note →