I architect reliable LLM infrastructure, evaluation frameworks that catch failures, and clinical-validated ML pipelines. I audit broken systems at scale and ship code that doesn't fail silently.
Currently scaling clinical ML at Hasprana Health Care Solutions. Previously debugged agentic systems at Lamatic.ai. Published research on LLM benchmarking (ICAC2N 2026).
Kairos — RAG Evaluation Infrastructure
MukhdaX — Face Provenance & Blockchain Verification
LocalBench — Offline LLM Benchmarking
RedOps — LLM Red-Teaming & Evaluation
Hasprana (Aug 2026 – Present)
U-Net retrain: 0.888 → 0.970 IoU (+9.2%), 70% error reduction. ONNX + INT8 quantization. 75% model size reduction. Clinical validation on 12 images + 251 tests.
Lamatic (Jan 2026 – May 2026)
Traced agent failures (hallucinations, routing errors, reasoning loops). A/B tested fixes. Deployed to production.
ICAC2N 2026 — Benchmarking Instruction-Tuned Small Language Models
Compared Phi-3-mini, Mistral-7B, Gemma-2. Finding: metric choice determines model ranking.
LangChain Contributions
- PR #31802 (merged) — Fixed missing
elsebranch in_evaluate_in_project() - PR #38465 (submitted) — KeyError fix in
file_toolcallback
Competitions
- Smart India Hackathon 2024 (National Finalist)
- Gen AI Exchange 2025 (National Finalist)
- EY Tecathon 6.0 (Technical Lead)
Code Auditing at Scale — 62K+ LOC analysis. God node detection. Coupling analysis. Architecture assessment.
Reliability Engineering — Kairos measures 12+ IR metrics with confidence intervals. RedOps generates forensic evidence. I measure failures, not just ship code.
Clinical Validation — Models validated on real patient data. Inference pipelines quantized, tested, benchmarked. Production-ready code.
Systematic Debugging — Trace agentic execution patterns. Identify hallucinations and routing errors. Fix via A/B testing.
Full-time AI/ML engineering roles (June 2027)
Focus: System design, evaluation rigor, production reliability, observable infrastructure.
Email — pradhananubhab25@gmail.com
LinkedIn — anubhabpradhan
GitHub — ANUBprad
Last updated: September 2026


