Ali Ameen

Helped underwriters catch risk combinations that manual review often misses in commercial property and auto insurance. Uses LLM structured extraction, deterministic risk rules, and anomaly detection, with low-confidence and flagged cases sent to human review.

Architected a public Agent-to-Agent (A2A) endpoint powered by an internal reasoning agent that autonomously dialogues with visiting client AI agents to discover needs, extract technical requirements, and issue dynamic project quotes based on live rate sheets. Enabled the internal agent to interactively negotiate service scopes with external AI buyers and auto-generate structured proposals.

Built onboarding and campaign creation as LangGraph agents, with every site action (campaigns, submissions, analytics, payments) available through chat via tool calls. Set up tiered moderation: cheap models for text, stronger ones for documents, a vision model for images and video.

Helped the firm run discovery across ~1,000 civil, tort, and criminal cases, with all discovery jobs rated 7.5/10. A case package (images, PDFs, text, user message, shortlisted case law) goes in, and relevant case law and judgments come back with relational data.

Helped cut recruiter screening from 4 hours to under 15 minutes per role. Built on FastAPI with per-recruiter data isolation: incoming CVs are deduplicated and merged, embedded in pgvector, and searched with BM25 + semantic hybrid search, with ASR voice queries and Google TTS replies.

Led ML and Unity engineers building real-time 6DoF bat and body tracking on low-end Android phones, no depth sensor. Owned the full ML pipeline: synthetic dataset creation, annotation, model training, and on-device inference.

Built predictive ML models with feature engineering on bowler and team analytics each IPL season to decode game plans, using Python, SQL, and pandas on structured historical match data. Ran live broadcast analytics for 200M+ viewers; the segment drew 1.5x the TRP of other analysis segments.

Shipped a farm simulation and a multiplayer paintball FPS with real-time networking on low-end hardware.

Led a team of four to ship a Homescapes-style renovation game in six months, cutting timelines ~20% across the group’s studios.

Built multiplayer VR gameplay for Curse of the Sunken Temple, using LEAP Motion hand tracking and holding a stable 90 FPS.

Organized medical courses and symposia in the UK. Coordinated with speakers, attendees and ensured smooth operations.

Built a multi-agent system turning 40–175 page deal documents into source-linked memos in 10 minutes instead of days. Cut the cost of one memo from $10,000+ in analyst hours to about $5, saving the firm $2.5M+ across 270+ documents.

B.S. in Electrical Engineering with focus on Computer Science and Electronics, on a full merit scholarship.