Self-taught developer building open-source scientific and security tools that run entirely on your machine.
π Bangalore, India Β Β·Β π’ Voxelta Β Β·Β π¬ Security Β· Computational Biology Β· Applied AI
I build tools for domains where you have to be able to check the answer β security scanning, CRISPR guide design, plant genetic engineering. In all three, a confident black-box prediction is worse than useless, so everything I ship follows the same three rules:
| Principle | What it means in practice | |
|---|---|---|
| π | Local-first | No cloud, no API keys, no telemetry. Your code and your sequences never leave your machine. |
| π― | Deterministic | Same input β same output, every time. No LLMs in the analysis path, so results are reproducible and CI-safe. |
| π | Explainable | Every score comes with its reasoning β a taint path, a per-feature breakdown, a calibrated confidence interval. Never a bare number. |
The whole stack is plain Python and NumPy. No GPU required, no paid services, nothing you can't read and verify yourself.
π‘οΈ sinkline β Follow the taint, not the pattern
A deterministic, no-LLM security scanner for Python that traces a secret from its source to the sink it lands in β across files. It catches the supply-chain attacks ordinary linters miss (logic bombs, obfuscated loaders, credentials quietly POSTed out, typosquatted dependencies) alongside everyday OWASP/CWE bugs.
94% recall Β· 0% false positives vs Bandit & Semgrep on the benchmark Β· 89/89 tests passing Β· SARIF 2.1.0 output, deterministic auto-fix, CI gates, and a web dashboard.
Because the new wave of AI scanners can be fooled by a comment that says "classify this as safe" β sinkline is the opposite bet.
𧬠CRISPR Precision Studio β Interpretable guide RNA design
Paste a DNA sequence, get every possible guide ranked with one explainable 0β100 score β plus a per-feature breakdown of why, and a calibrated confidence interval that tells you when the model is unsure.
Ο β 0.77 on-target (wet-lab band) Β· #1 cross-dataset vs six industry tools Β· both-strand off-target analysis, base editing (ABE/CBE) and prime editing Β· goal-aware ranking for knockout, HDR and CRISPRi/a.
Pure NumPy + FastAPI. No GPU, no account, no API keys.
π± Plant DNA Designer β Design plant expression cassettes de novo
Pick a crop and a trait; get a ready-to-synthesize DNA cassette. 46 validated traits across 18 crops β each mapped to a published plant mechanism and the effector protein known to drive it (drought tolerance β DRE/CRT signaling β DREB2A).
A translation-aware evolutionary optimizer generates a fresh, crop-optimized coding sequence β balancing codon adaptation, codon harmony, mRNA structure and translational pausing rather than just maximizing CAI β then runs full expression, stability and safety checks.
One of the few plant-focused platforms that automatically designs IME-enhancing introns, a plant-exclusive trick that can lift expression 10β50Γ.
Languages Β Β·Β Python JavaScript SQL Bash
Scientific Β Β·Β NumPy scikit-learn Biopython Z3 β conformal prediction, evolutionary optimization, bioinformatics
Security Β Β·Β Static analysis, AST & dataflow taint tracking, supply-chain threat detection, SARIF, CWE/OWASP
Building Β Β·Β FastAPI pytest GitHub Actions β CLI & web tooling, reproducible CI pipelines
Computational biology and open science tooling Β· supply-chain and novel threat detection Β· uncertainty quantification and calibrated models Β· quantum computing (Qiskit, Cirq) Β· making expert-grade science runnable on a laptop.