name: Vinisha Bhagwani
role: AI/ML Engineer — RL, Agentic Systems, Full-Stack
education: B.Tech CSE @ SPIT '27 • Minor MBA @ SPJIMR '27
location: Mumbai, India
focus: Reinforcement Learning • Multi-Agent Systems • Reward Engineering9× Winner • 5× International Finalist • Top 2.6% globally @ Meta Hackathon
| 🏅 | Event |
|---|---|
| 🥇 | IIT Delhi National |
| 🥇 | BITS Pilani |
| 🥇 | CodeKong International |
| 🥇 | BrainRot International |
| 🥇 | DotDotCom International |
| 🥇 | NextAI International |
| 🥇 | NMIMS • Hacknomics • UI/UX |
| 🎖️ | Meta Hackathon — Top 800 / 31,000+ |
| 🎖️ | GenAI Zürich |
| 🎖️ | HERE Technologies |
| 🎖️ | Google Agentic Cloud |
| 🎖️ | Google GenAI |
|
Closed-Loop DBS Reinforcement Learning Environment OpenEnv-compatible RL benchmark for adaptive Deep Brain Stimulation in Parkinson's patients. Calibrated against the peer-reviewed Fleming et al. (2023) biophysical model from Journal of Neural Engineering. ▸ 3-task curriculum with continuous DBS action space (amplitude, pulse width, motor command) ▸ 8-component reward balancing motor function, beta/tremor suppression, safety budget, and clinical smoothness ▸ Bilinear-interpolated entrainment surface from a 12×15 parameter sweep ▸ Patient-profile dynamics modelling fatigue, recovery, side-effect sensitivity ▸ Deployed publicly on HF Spaces with deterministic graders
🥈 Meta Hackathon Finalist — Top 800 / 31,000+ (2.6%) |
RL Benchmark for Software Delivery Under Uncertainty Deterministic OpenEnv environment that frames project management as a sequential decision problem — agents make daily staffing, prioritization, and intervention calls under dependency graphs and burnout cascades. ▸ Dual reward system: 8-component dense step rewards + multi-dimensional terminal grader (completion, deadline, budget, team health, satisfaction) ▸ Anti-exploit penalties for thrashing and idle-employee farming ▸ Benchmarked Qwen 2.5-72B across 3 difficulty tiers ▸ Baseline scores 0.91 / 0.65 / 0.30 on the 22-day, 6-disruption hard scenario — exposes a clear optimization frontier
|
|
Agentic Tactical Decision System Multi-agent reasoning pipeline orchestrating specialist agents through a stateful LangGraph workflow. Won IIT Delhi National over 200+ teams. ▸ Hierarchical agent topology with shared scratchpad memory ▸ Tool-calling layer with structured output validation ▸ Real-time decision streaming via Server-Sent Events
🥇 IIT Delhi National — 1st Place |
Multi-Agent Competitor Intelligence Platform End-to-end pipeline for autonomous market-research with human-in-the-loop checkpoints at high-stakes inference steps. ▸ Hybrid source-tagging system separating verified facts from inferred claims ▸ Parallel agent fan-out for source coverage, then consensus-merge ▸ HITL review gates before report finalization
🏆 Built for CodHer'26 |
|
Geospatial AI for Map Data Quality Automated verification pipeline for Points-of-Interest data combining unsupervised geospatial clustering with adversarial multi-agent reasoning. ▸ Density-based clustering for spatial anomaly detection ▸ Pros vs Cons dual-agent debate framework with structured rebuttals ▸ Softmax decision layer for calibrated confidence scoring
🎖️ HERE Technologies Hackathon — Finalist |
Agentic Learning Platform Personalized education platform that generates, evaluates, and adapts learning material from a single learner-profile prompt. ▸ Curriculum generation with concept-dependency awareness ▸ Adaptive difficulty via outcome-conditioned regeneration ▸ Production deploy on Vercel with edge functions
🎖️ Google GenAI Hackathon — Finalist |
🔬 Production-grade Spring Boot architectures
└─ deep dive: distributed job schedulers (fintech + platform domains)
🤖 Agentic AI for academic topic modelling
└─ LangGraph + BERTopic + MistralAI → Gradio on HF Spaces
Team Lead & Fullstack Engineer (Research Intern) — AiM4u Software Solutions Jan – Jul 2025
Led 5-engineer team as a 2nd-year intern; shipped Gyan (internship-discovery platform) zero-to-prod in 6 months on Next.js + Prisma. Personally rebuilt the auth and transactional mailing pipeline that had blocked launch for months.
Freelance Fullstack Engineer — Credge Education Jul – Oct 2025
Sole engineer on a paying-customer SaaS. Scoped, architected, and shipped 3 production modules (counsellor workflows, editorial pipelines, program matching) end-to-end across 3 user roles with no senior oversight.
Backend Engineering Intern — Squad Infotech Jun – Jul 2023
Reduced multi-week manual scheduling to seconds via a constraint-based Java algorithm resolving conflicts across classrooms, credit limits, and faculty availability. Delivered as a 1st-year intern after the prior approach had been abandoned.
"Building agents that learn, systems that scale, and teams that ship."
