← Ruhan Sahasi
Ruhan Sahasi
LinkedIn
Education
University of PennsylvaniaPhiladelphia, PA
B.S. Computer Science + Accelerated M.S. Computer Science (Machine Learning & Robotics)Expected May 2028
Relevant Coursework: Data Structures, Compilers, Multivariable Calculus, Linear Algebra & Differential Equations, Statistics
Professional Experience
Marlo (a16z SR004)San Francisco, CA
Engineering InternMay 2026 - Aug 2026
- Built and shipped Leonardo, Marlo's model-improvement feedback loop that turns managers' edits of AI-drafted emails into diagnosed, routed product fixes; merged about 30,000 lines of reviewed Python into production across 6 PRs.
- Ran Leonardo weekly on live production data (about 1,000 draft/sent pairs across 32 talent managers), generating 59 statistically gated recommendations at a 56% acceptance rate; about 35 accepted fixes shipped in three weeks.
- Closed the loop with a human-review surface: weekly accept/reject decisions sync into suppression and gold sets, while a pre/post outcome panel measures whether shipped fixes changed edit rates.
AMDPhiladelphia, PA
EPYC CPU Product AnalystSep 2025 - May 2026
- Sized EPYC demand across priority workloads - virtualization, customer analytics, and AI inference - building a top-down and bottom-up TAM/SAM/SOM model for cross-verification ($14.8-$20.3B in 2026) and segmenting each workload by core count, memory bandwidth, and accelerator-offload requirements.
- Mapped user behavior through mixed-method research with 10+ Chief Procurement Officers and hundreds of industry experts to quantify buying criteria across Zen-architecture EPYC platforms: core density, memory channels, PCIe/CXL I/O, performance per watt, and TCO versus Intel Xeon.
CoreThink (Stealth AI Startup Built for Neuro-Symbolic LLM Orchestration)San Francisco, CA
Software Engineering InternMay 2025 - Aug 2025
- Built and maintained a Python evaluation pipeline to benchmark 11 models (7 base, 4 instruction) on reasoning, code, tool use, and multilingual tasks; wrote reports that helped the founders identify performance gaps.
- Analyzed 12,000 user-agent interactions across three stealth test environments; shipped prompt and routing changes that increased target-task accuracy, improved first-response helpfulness, and reduced latency by 23%.
Projects
Leonardo - Edit-Diagnosis & Routing Engine for AI Email DraftsSan Francisco, CA
- Built a two-agent LLM pipeline (Gemini 2.5 Pro on Vertex AI): normalized Levenshtein scoring over ClickHouse draft/sent pairs routes heavy rewrites into a diagnosis agent with five fix routes and a pattern-grouping aggregation agent.
- Added Wilson 95% lower-bound promotion gates on a cross-week pattern registry, fail-closed idempotent weekly folds, immutable Cloudflare R2 snapshots with compare-and-swap commits, and admin-gated FastAPI endpoints.
UIUC Bustle - Bus Tracking ApplicationPhiladelphia, PA
- Built a full-stack transit system with React Native/Expo Router, FastAPI, and SQLite that merges MTD real-time, GTFS, and OSM/Overpass/Nominatim APIs into a multi-leg route engine with live map rendering and vehicle overlays.
- Integrated the Claude API for LLM reranking, natural-language explanations, and context generation.
- Built background scheduling and notifications in TypeScript with idempotent cancel-then-reschedule logic, TTL caching, and cache coherence between AsyncStorage and live API reads.
Leadership & Community Impact
Veterans History Studio (National Nonprofit for Veteran Support)Palo Alto, CA
FounderSep 2022 - Present
- Founded a national nonprofit that conducted 500+ veteran interviews, helped drive a $14M county allocation, secured a $10K/month Google grant and $12K in fundraising, and built VetBot, an ML assistant serving 30,000 veteran queries.
Penn Computer Science SocietyPhiladelphia, PA
Vice President of Internal OperationsSep 2025 - Present
- Manage events for 100+ club members, including fireside chats, technical showcases, and hackathons.
Additional Information
Skills: Programming Languages: Python, C++, JavaScript, Java | Systems: Docker, Git, AWS | Data/Backend: SQL, REST APIs, FastAPI | Machine Learning: ML libraries (TensorFlow), Claude Code, LLM APIs
Awards: Presidential Service Award (Gold - 3x), National French Competition (Gold - 3x), National Debate Honors (Top 250 in Public Forum, 2x Tournament of Champions Qualification), $10K/month Google Grant