Aditya Bisht
Building AI systems, agents, and products at the intersection of research and real-world scale.
01 — About
Who I am.
I'm a third-year Computer Science student at Northeastern University (AI concentration, 3.52 GPA), graduating May 2027. My work sits at the intersection of AI systems, agent infrastructure, and data engineering — I care about building things that actually run in production.
I founded GatePoint — a production security gateway for AI agents with a cryptographic trust chain, policy decision engine, and a zero-change TypeScript SDK. It placed 2nd at the Husky Startup Challenge. Before that, I built sentiment pipelines and admin tooling at Rezolve.ai and KYC automation infrastructure at Zolve.
Beyond work: Z Fellows alum, member of MIT Sundai Club, and 1st-place finisher at HackBeanPot 2025 with Memora. I like hackathons — the constraint is the point.
Long-term, I'm drawn to AI safety and interpretability. The systems we're building are getting powerful fast, and I think getting the foundations right matters enormously. Anthropic is where I want to end up.
Tech Stack
02 — Experience
Where I've worked.
Allowance (Y Combinator)INCOMING
AI Product Engineering Co-op
July 2026 – Dec 2026
- Developing payment infrastructure that programmatically issues one-time virtual cards with approval controls, spending limits, and merchant restrictions to authorize purchases on behalf of users.
- Designing and shipping features and internal tooling across frontend and backend to power autonomous purchasing workflows (travel booking, restock tracking, price-drop execution).
- Building evaluation systems for transaction reliability and checkout success rates; testing payment flows to surface and resolve edge cases before production deployment.
GatePoint
Founder & Lead Engineer
Sept 2025 – Feb 2026
Boston, MA
- Architected a production AI agent security gateway — Express + PostgreSQL (15-model Prisma schema), React/Vite/Tailwind dashboard, published TypeScript SDK — enabling enterprises to enforce security policies and audit every tool call made by AI agents.
- Designed a 3-layer cryptographic trust chain (Ed25519 challenge-response, 15-min JWTs, CA-signed certificates with token-nonce binding) with per-request RFC 8785 canonical JSON signatures — delivering 3 independent cryptographic verifications and 2 replay-prevention layers per tool call.
- Built a policy decision engine: 18 JSONPath condition operators, 4 effects, 3 evaluation phases, and GPT-4 Turbo policy generation with Zod validation as a hard correctness gate.
- 2nd place, Husky Startup Challenge.
Rezolve.ai
Software Development Engineering Intern
July 2025 – Dec 2025
Boston, MA
- Built real-time sentiment analysis pipeline (OpenAI + Anthropic APIs) for customer support transcripts with streaming inference.
- Implemented Redis caching for API prompts/responses, reducing AI platform costs by 50% and improving response latency by 2x.
- Built React admin dashboard for DeskIQ with authentication, master account controls, and test coverage for user management.
Zolve
Data Engineering Intern
Jun 2024 – Sept 2024
Bangalore, India
- Built automated KYC verification pipeline (MTCNN, YOLO, ArcFace) processing 15,000+ documents at ~92% accuracy.
- Optimized inference throughput ~30% through batching and GPU utilization improvements; built evaluation metrics and failure analysis dashboards.
03 — Skills
Tech Stack.
Languages
AI / ML
Web & Backend
Data & Infrastructure
04 — Projects
What I've built.
GatePoint
AI Agent Security Gateway
Production security gateway for AI agents — cryptographic trust chain, policy decision engine with 18 operators, GPT-4 policy generation, transparent SDK with zero agent code changes.
Neocortex
Semantic Code Intelligence
Parses GitHub repos into AST entity graphs at ~1,000 entities/sec, generates OpenAI embeddings, returns ranked token-budgeted context to any LLM in <500ms.
ImageDoctor
Vision Transformer IQA System
Full-reference image quality assessment on KADID-10k (10,125 images). 50.6% validation loss reduction over 50 epochs. ~70% GPU memory reduction via selective ViT freezing + gradient accumulation.
Loan Default Prediction
ML Credit Risk Modeling
1.35M Lending Club loans, 0.72 AUC-ROC with XGBoost. All feature engineering in SQL (DuckDB). SHAP interpretability. Full error analysis revealing systematic failure patterns by loan grade.
Marketing A/B Analysis
Statistical Experiment Analysis
588K-user A/B test. 43% relative conversion lift confirmed across 3 statistical methods. Dose-response curve, Simpson's paradox detection, Bonferroni-corrected timing analysis.
Memora
Alzheimer's Remediation App
AI-generated memory quizzes from tagged photos, TikTok-style feed, drag-and-drop annotation. Built in <36 hours.
FUSE Filesystem
Systems Programming
1MB FUSE-based filesystem in C — file creation, deletion, renaming, hierarchical directories, byte-level read/write on a custom block storage layer.
05 — Contact
Let's build
something.
Open to co-ops, research collaborations, and interesting problems. If you're working on something ambitious, I'd love to hear about it.