01
Aug 2026 — PresentPalo Alto, California
Founding Member of Technical Staff
Life Sciences Startup
Building data infrastructure for AI-driven scientific discovery.
Builder.
MadhavChauhan.
Founding Member of Technical Staff at a life sciences startup.
I build the infrastructure around intelligent systems: the data, evaluation, and product layers that make them useful.
Professional experience
Close to the hard questions.
Early teams, real technical ownership, and work that moves from an ambiguous question to a dependable system.
02
Dec 2025 — May 2026Palo Alto, California
Software Engineer, AI/ML
Stealth-stage AI Company
Worked on proprietary foundational AI focused on the mechanisms that shape model behavior and improve model reasoning.
Earlier
Machine Learning Intern
Haystream
Built credit-risk modeling, scoring, and portfolio analysis workflows.Selected public work
Two things worth opening.
Not a project dump. These are the public artifacts that best show how I investigate, build, and evaluate technical systems.
01Mechanistic interpretability · Research engineering
MicroScope
A reproducible interpretability toolkit that tests whether sparse model features carry real, causal structure—not merely compelling explanations.
What held upGemma 2 · verified experiments
01
Reproduced Gemma Scope and SAEBench baselines before extending them.
02
Found a sparse 5–10 feature circuit that preserves 94–97% of full accuracy.
03
Recorded every run, seed, hardware choice, and cost—including inconclusive results.
- Python
- Gemma 2
- SAEs
- Modal
- Evaluation
02Open-source ML tooling
cortex-score
A Python package for scoring video against predicted engagement across five cortical networks, built on Meta FAIR’s TRIBE v2.
- Five network-level scores
- Stable CPU post-processing
- Versioned, machine-readable output
Open source
Work in other people’s codebases.
A public trail of reading unfamiliar systems, finding the edge case, and leaving the code better than I found it.
01
Roboflow / Supervision
Merged dataset format support, windowed GeoTIFF reads, and correctness fixes across a widely used computer-vision toolkit.
6 merged pull requests02Roboflow / Inference
Corrected aggregation behavior in the production workflow engine’s Data Aggregator block.
Merged pull request #2388About
I have experience building systems at the intersection of AI, infrastructure, and product.
I am an early-career engineer already working on difficult AI, infrastructure, and product problems.
My work sits at the seams: research systems that need dependable infrastructure, backend platforms that need a thoughtful product surface, and ML pipelines that have to support real decisions.
I'm from Houston, completed a B.S. in Computer Science at The University of Texas at Dallas in May 2026, and now work in Palo Alto.
- Education
- B.S. in Computer Science
- University
- The University of Texas at Dallas
- Completed
- May 2026
Direct line
Let's make
something useful.
The best way to reach me is by email or LinkedIn. GitHub has the implementation details.