Lastline | PR → QA Review Video
Turns pull requests into review-ready evidence. Generates a guided QA walkthrough of the actual product experience, stitches it into a single review video, and routes feedback back into the team loop.
AI/ML Engineer
Founding Engineer · OrphaNova
NYC/SF
Hi I'm Prashant Soni
I build and ship AI systems end-to-end, from idea to production.
Focused on LLM systems, agents, and real-world deployment.
Recent Ships:
SpaceGuard: Kalshi for space events 🚀. A predictive intelligence platform providing satellite collision risk assessment via real-time market signals.
Homecastr: FaceTime for real estate 🏠. A conversational video AI that transforms static property listings into interactive, agent-led experiences.
OrphaNova: The OS for orphan drug R&D 🧬. An AI-native research engine architected to accelerate therapeutic discovery for rare diseases.
Lastline: Lastline takes a pull request and turns it into a reviewable QA video for the team. PR → QA → Video → Chat → Feedback
Experience
Founding Engineer & Lead Developer · OrphaNova
Building OrphaNova, an AI research system for rare disease discovery. Designed a 7-stage pipeline that retrieves biomedical literature, extracts disease-gene relationships, predicts protein structures (AlphaFold2), screens drug candidates (RDKit), performs molecular docking (Chai-1), and generates publication-ready research reports.
Research Assistant · Yeshiva University
Developed deep learning models for cardiomegaly detection from veterinary X-rays, achieving 85.75% accuracy on 2,000+ images. Built OrphanAtlas, a rare disease research platform combining GPT-based retrieval pipelines with biomedical datasets, now used by 500+ researchers.
Received Outstanding R&D Award at the Yeshiva University Research Symposium. Featured in YU News
Projects
Featured builds, shipped and live
SpaceGuard | Satellite Collision Risk Intelligence (Kalshi for Space)
Real-time satellite collision risk intelligence platform: think Kalshi for space events. Tracks 800+ satellites, live prediction markets, critical portfolio risk ($2.4B at risk), and active hedges. Built for space industry professionals and prediction market enthusiasts.
Other Projects
Kalshi Continual Learning Trading Bot
Event-driven trading agent for Kalshi prediction markets. Learns from new market data in real-time, generates trading signals, and executes trades with built-in risk controls and backtesting framework.
Agents4Science
Agent-based generative AI workflows for drug repurposing research. Mines biomedical literature, extracts evidence, and generates hypotheses. Built with LLM + RAG pipelines and domain-constrained prompting to reduce hallucinations in sparse data.
Multimodal Stock Price Predictor
Deep learning pipeline combining LSTM time-series analysis with FinBERT sentiment from news/social media. Achieved 15% higher accuracy than single-modality models. Includes backtesting and financial metrics (Sharpe ratio, MSE, RMSE) for risk-adjusted insights.
Kaggle ConnectX RL Agent
Built reinforcement learning agents using Q-learning, Deep Q-Networks, and Monte Carlo Tree Search. Implemented epsilon-greedy exploration and policy gradients. Secured 7th place globally (top 0.5%) through hyperparameter tuning and reward shaping.
GitHub Highlights
Latest repositories from GitHub.
Education
Research & Publications
Top publications from ResearchGate.
Visit ResearchGate ProfileTechnical Skills
Core
Python, PyTorch, TensorFlow, FastAPI
AI Systems
LLMs, RAG pipelines, agent systems, embeddings, multimodal, ClawAgents
Infra
APIs, vector DB (FAISS), basic cloud (GCP/AWS)
Schedule a Call
Pick a date and time below to book directly.
Contact
Phone
+1 551 362 9944Location
NYC / San Francisco