Devansh
Baviskar

Super-intelligent agent orchestration system using ASI:One to autonomously select, route, and execute Hugging Face models with transient/persistent runners and MeTTa knowledge graph.

• Built a super-intelligent agent orchestration system using ASI:One to autonomously select, route, and execute Hugging Face models.
• Implemented transient vs. persistent model execution with subprocess-based runners, reducing cold-start latency.
• Designed a MeTTa-based knowledge graph to infer model capabilities, track usage frequency, and keep high-demand models warm.
• Engineered an Agentverse architecture with address-based messaging and secure sandboxed execution across HF Agents.

Nationalized platform for onboarding patients & medical officials. Wallet-based report storage, multi-factor KYC, on-chain deployment on Aptos blockchain with Next.js & Octa auth.

• Ideated a Nationalized Platform for on-boarding patients and medical officials likewise for ease of access to healthcare services.
• Proposed decentralized wallet-based report storage, Multi-factor KYC, On-chain deployment for Privacy and Security.
• Integrated Aptos blockchain with Next.js framework with Octa authentication to implement bridge between Web2 and Web3.

Event-driven RAG ingestion pipeline via FastAPI and Redis Streams achieving 1,000+ docs/sec throughput, Qdrant vector search, and 100% test coverage with AsyncMock.

• Architected an event-driven RAG pipeline via FastAPI and Redis Streams, achieving 1,000+ docs/sec throughput with zero data loss.
• Eliminated duplicate worker tasks across scaled nodes by engineering custom distributed Redis TTL locks for strict concurrency.
• Cut vector ingestion latency by >60% using Qdrant and SentenceTransformers with asynchronous queue decoupling.
• Enforced 100% test reliability with zero Mypy/Ruff errors by authoring 36 automated unit tests using AsyncMock.

Offline-first Next.js ledger platform with face_recognition + OpenCV biometric login, embedding-based auth, audit timeline tracking, and live support for 50+ users.

• Built an offline-first Next.js ledger platform with local persistence, customer record management and Excel export workflows.
• Engineered biometric login using face_recognition and OpenCV with embedding-based authentication and session authorization.
• Implemented audit timeline tracking and biometric-secured access control for a live production deployment supporting 50+ users.

Multimodal AI DAW converting video & text to multi-track audio using Gemini Vision, 2-stage FAISS vector search, async PyTorch batching, and custom Python DSP mixing.

• Engineered an end-to-end multimodal AI DAW with React 19 and FastAPI that converts video and text into multi-track audio using Gemini Vision beat-sheet extraction.
• Architected a 2-stage FAISS vector search engine with a 0.72+ confidence gate for semantic background music retrieval using bge-large embeddings.
• Optimized backend generation throughput via category-based async PyTorch batching, reducing pipeline API round-trips by 60%.
• Designed a custom Python DSP mixing engine executing dynamic sidechain audio ducking, gain keyframing, and peak normalization synced with a WaveSurfer.js timeline UI.

Full-stack progressive social platform with TypeScript Jaccard affinity engine, TanStack Query progressive disclosure, Supabase SSR middleware, and Ionic Capacitor native bundles.

• Architected a full-stack progressive social platform leveraging Next.js 14 App Router, TypeScript, and Supabase/PostgreSQL with Drizzle ORM, shipping native iOS and web bundles via Ionic Capacitor.
• Engineered a custom multi-dimensional affinity engine in TypeScript using Jaccard similarity matrices to output real-time match scores (0–100%) at sub-100ms query latency.
• Designed a progressive disclosure interaction pipeline with TanStack Query and Zustand, tracking reciprocal message exchange milestones (1–15+ turns) to algorithmically unblur user profiles.
• Built high-throughput trust & safety architecture using Supabase SSR middleware, custom Content Security Policy (CSP) headers, and JSONB audit snapshots for real-time moderation.

LangGraph-orchestrated coding agent for analyzing codebases and proposing streamed refactor diffs for protocol integrations over Yellow SDK and Nitrolite within a Next.js Monaco workspace.

• Built a LangGraph-orchestrated coding agent to analyze codebases and propose streamed refactor diffs for protocol integrations.
• Engineered RAG-based code generation over Yellow SDK and Nitrolite using vector search for state-channel workflow scaffolding.
• Developed a Next.js Monaco workspace with diff streaming, agent routing and terminal for protocol-aware code transformation.

Base L2 creator funding platform using Next.js and Solidity featuring Quadratic Funding, hybrid Firebase/IPFS metadata architecture, and on-chain voting contracts.

• Architected a Base L2 funding platform using Next.js and Solidity for transparent community governance.
• Engineered a Quadratic Funding mechanism to ensure fair reward allocation based on unique contributor count.
• Built a hybrid Firebase/IPFS architecture for rich metadata, optimizing profile browsing and frontend speed.
• Developed on-chain voting contracts for verifiable ranking and automated payouts at the end of funding rounds.

TypeScript VS Code extension integrating SQL/dbt parsing, OpenMetadata APIs, schema hovers, lineage visualization, SQL autocomplete, CodeLens, and local Docker testing stack.

• Built a TypeScript VS Code extension integrating SQL/dbt parsing, OpenMetadata APIs, schema hovers and metadata operations.
• Engineered lineage visualization, SQL autocomplete and CodeLens tooling using parser logic, OM client resolution and lineage APIs.
• Developed core extension modules for metadata-aware workflows across SQL and dbt, supported by a local Docker testing stack.