Work.

End-to-end AI stack combining a custom UI layer (Knowledge Graph, ChatAlphaMeld, Bioinformatics, Literature Analysis, AlphaClinMeld) with 12 ML models — including NER, RE, evidence scoring, digital-twin CT predictors and landscape assessment — running on PostgreSQL + Neo4j and self-hosted LLM APIs.

Guided drug-discovery hypothesis-generation platform built on the AlphaMeld Knowledge Graph. Supports four starting points — disease-first, target-first, indication expansion, and combination therapy — through a step-by-step workflow (project → config → exploration → agent scoring → ranking → hypotheses). A project-aware Discovery Copilot accompanies every step. Multi-tenant SaaS with SSO, RBAC, and audit logging.

Two agentic workflow engines fine-tuned on biomedical tasks; outperformed leading general-purpose frontier LLMs; delivered richer step-by-step rationales critical for complex clinical questions.

End-to-end AI stack combining Knowledge Graph, Clinical Trial Prediction, Literature Analysis, Patient Digital Twin, TPP, TOP with 12 ML models running on PostgreSQL + Neo4j and self-hosted LLM APIs.

Clinical-trial–centric SaaS module on AlphaMeld® graph analytics; benchmarks pipeline assets and quantifies comparative risk. Adopted by two BD teams within six months, standardizing evidence-based indication-prioritization across 200 active programs.

Outcome-focused digital-twin framework built on the EUSTAR systemic sclerosis registry — longitudinal records for 23,000+ patients across 200+ centers. A PyTorch deep/generative stack (a mask- and time-aware GRU-D sequence forecaster plus a conditional sequence-VAE twin generator) produces prognostic scores and synthetic placebo trajectories for organ-specific and composite clinical outcomes. Designed as an ML-generated digital control arm — a PROCOVA/ANCOVA covariate that lifts trial power and cuts sample size — aligned with FDA digital-twin and EMA/ICH MIDD guidance.

Conditional variational auto-encoder (CVAE) pipeline that synthesizes patient-level digital twins for Phase II Type 2 Diabetes trials. Generated matched placebo trajectories and predicted drug-response curves — proving feasibility of in-silico arms for protocol optimization and hypothesis testing.

Full-stack platform that collects, validates, and structures biomarker market intelligence — pricing, availability, and distributors — for neurological diagnostic kits worldwide. A six-layer, event-driven multi-agent pipeline pairs deep web research with schema-enforced LLM synthesis; every data point carries source lineage, a credibility rating, and a stated-vs-extrapolated confidence flag. Vision-first PDF extraction, resumable checkpointing, and a canonical-run batch model keep large refreshes auditable. Built on Next.js, Drizzle/PostgreSQL, and Inngest for event-driven orchestration.

External-innovation asset-screening dashboard on the AlphaMeld platform. A disease-first funnel narrows the ~9,000-disease rare-disease universe into a ranked longlist and a focused asset shortlist, with provenance on every score, gate, and number — deterministic and computed values kept visibly distinct from LLM-proposed-then-confirmed ones. A React 19 + TypeScript front end runs over a Hono API and PostgreSQL + pgvector for similarity search and entity resolution. Demonstrated on synthetic, illustrative data.

Investment research platform for small/mid-cap NASDAQ biotechs where a single trial readout can move the stock 50%+. Syncs clinical-trial data (ClinicalTrials.gov, Clinical Navigator) with live stock and financials (Alpha Vantage), then runs LLM catalyst discovery and streaming AI predictions for upcoming readouts, PDUFA dates, partnerships, and conference events. Includes bulk catalyst briefings and an interactive research chat.

Secure, role-based platform for pharma/biotech regulatory affairs teams to search, analyze, compare, and manage global drug and biologic approvals. Covers pharmaceuticals, biologics, biosimilars, and generics across 12+ regulators. Combines role-based dashboards, pathway and classification guides, dossier preparation support, a centralized guidelines library, Q&A with expert responses, and LLM-powered document analysis.

Fully agent-orchestrated, multi-step LLM process. Input parameters, run a series of LLM processes to extract information from the web, compile, and store in a dashboard or export to Excel.

End-to-end booking and scheduling platform — custom-built to client specification with real-time availability, notifications, and management dashboard.

AI assistant that mines global regulatory updates and auto-drafts submission dossiers; established a new non-pharma revenue channel.