~/thesisSWE · ML ENGINEERING

I build systemsyou can look inside.

Software engineer working on ML systems. UCLA, B.S. Computer Science and Mathematics. Most recently a founding engineer at Ryniant; before that, interpretability research in UCLA’s NLP lab.

LOC Los Angeles, CAGRAD June 2026OPEN new grad SWE · ML

WORK

[01]Selected workALL FIVE →

01
May 2026 — July 2026
Ryniant
Founding Engineer · AI drug-synthesis platformRebuilt a synthesis-planning product so that nothing reaches a chemist unless it has been computed or found in the literature.
- Python
- FastAPI
- RDKit
- AiZynthFinder
- LangGraph
- FAISS
100% of displayed values structure-computed or literature-verified+25% discoverable synthesis routeshours → minutes synthesis planning timeRead →
02
September 2025 — May 2026
UCLA NLP Lab
Machine Learning Researcher · LLM interpretabilityLocated the components inside open-source language models responsible for hallucination, then intervened on them to prove the relationship was causal.
- PyTorch
- TransformerLens
- NNsight
- SAELens
- GPT-2
- Llama
6 / 32 layers carrying most reasoning-relevant signal3 attention heads necessary for hallucination behavior16 monosemantic features tied to hallucination-prone reasoning−30% hallucination rate under steering interventionRead →
03
June 2025 — September 2025
Genter Capital Management
Data Analyst Intern · Wealth management firmAutomated an IRS-mandated compliance calculation across 200+ client accounts, where the cost of a wrong number is a federal penalty.
- Python
- Pandas
- Plotly
- Schwab API
- BeautifulSoup
200+ client accounts under automated RMD calculation~40 hrs saved during compliance season3+ hrs/wk cut from portfolio reportingRead →
04
January 2025 — June 2025
Chelle AI
Machine Learning Engineer · LLM-based learning platformReplaced gut-feel prompt engineering with a rubric-based evaluation framework and automated scoring loops.
- Python
- LangChain
- FastAPI
- OpenAI API
- Google Gemini
−25% hallucination rate after rubric-driven prompt selection+18% student engagement scoresRead →

[02]Built independently

Sentinel

VIEW SOURCE →

Agentic evaluation of live web apps, with a verifier that measures its own false-positive rate.

Five subagents drive real user journeys through a live web app and propose defects. A deterministic verification layer then decides which claims are real before anything reaches the report — every finding has to carry a replayable step sequence and one machine-checkable assertion, or it gets dropped.

The self-measurement is the point. Against a build with eight seeded defects: 15 raw agent claims, 7 published, 2 suppressed, a 0% false-positive rate against 14.3% for the same agents with the verifier removed, and 7 of 8 seeded defects found. The repo also ships an honest status doc on the six things it would need to actually be a product.

JavaScript
Playwright
Claude agents

0%false-positive rate

14.3%without the verifier

7/8seeded defects found

[03]Why it is all the same job

Opening the box

The work I keep coming back to is making opaque things inspectable. In UCLA's NLP lab that was literal: finding which layers and attention heads carry a behaviour, then steering them to prove the link was causal and not coincidental. At Ryniant it meant a synthesis route a chemist could trace to a citation instead of taking on faith. At Chelle it meant a rubric, so a prompt decision became something you could run an experiment on rather than argue about. At Genter it meant a required distribution a client could actually be shown, deadline attached, instead of a number living in someone's spreadsheet.

A system you cannot inspect is one you can only trust or abandon. Most of what I have built exists to give people a third option.

[04]Contact

Say hello

Open to new grad and early-career software and ML engineering roles. The fastest way to reach me is email.

aadim@ucla.edu linkedin.com/in/aadi-mala github.com/Aad-im résumé.pdf

I build systemsyou can look inside.

Ryniant

UCLA NLP Lab

Genter Capital Management

Chelle AI

Sentinel

Opening the box

Say hello