Submitted by: Saurav
Date: 16 October, 2025
Amount Requested: 30000 USD
Project Duration: 3 Months + 9 Months operational costs.
Contact: saurav@bytebell.ai, bytebell.ai
Summary
Have you ever tried to understand how zero-knowledge proofs actually work? Or attempted to write a proof in Groth16? If you have, you know the feeling. Drowning in academic papers, scattered GitHub issues, outdated documentation, and forum threads that reference code from three versions ago.
Learning Zcash’s technology stack means navigating through multiple worlds at once. You start with the whitepaper, jump to a GitHub discussion, check the docs (which may or may not be current), ping someone on Discord, and still end up confused about which cryptographic primitive you should actually be using. It’s genuinely difficult, not because the technology is poorly designed, but because the knowledge about it lives everywhere and nowhere at the same time.
This isn’t just inconvenient. It’s a barrier to adoption.
Zcash’s technology is genuinely groundbreaking. Zero-knowledge proofs, shielded transactions, and privacy-preserving protocols represent some of the most important innovations in blockchain. But here’s the harsh truth: if developers can’t understand it, they won’t build with it. And right now, understanding Zcash means piecing together knowledge from multiple GitHub repositories (zcashd, zebra, librustzcash, halo2), academic papers spanning years of cryptographic research, documentation that doesn’t always reflect the latest protocol changes, forum discussions with critical context buried 200 posts deep, Discord channels where answers live and die in real-time conversation, and blog posts explaining concepts that reference code that’s since been refactored.
A new developer trying to build on Zcash might spend 30% of their time just hunting for information that already exists somewhere in this ecosystem. Experienced contributors constantly re-explain the same concepts. Critical knowledge walks out the door when contributors move on.
This is information entropy in action. And for a technology as complex as Zcash, it’s not just inefficient. It’s an existential threat to growth.
What Bytebell Does for Zcash
Imagine a developer asks: “How do I generate a shielded transaction using Orchard?”
Instead of spending hours searching through repos, docs, and forums, they get an instant answer with complete context. The relevant code from the current version. Links to the specific documentation sections. References to the forum discussions that shaped the design decisions. Examples from other projects that implemented similar patterns. Even the historical context of why it works this way.
Bytebell becomes the single platform that can explain everything about Zcash to any developer, at any skill level.
How Developers Access It
MCP Integration
We provide Model Context Protocol servers that plug Zcash’s knowledge directly into Cursor, Claude Desktop, and any MCP-compatible IDE. When you’re writing code, you get completions based on actual Zcash implementations, not generic suggestions. Ask a question and get answers pulled from current docs and real repositories. The context lives where you work.
Website Widget
An embeddable chat widget for z.cash docs, GitHub READMEs, and community forums. Developers can ask follow-up questions without leaving the page. No more switching between tabs to understand a code sample. The knowledge shows up exactly where people need it.
WebChat [https://zcash.bytebell.ai]
A full web interface for developers who want to explore deeply. Trace how a cryptographic decision evolved over time. Understand how different concepts connect. Navigate Zcash’s complete technical history through conversation instead of archaeology.
Chrome Extension
Highlight any code, documentation, or discussion anywhere on the web and instantly get context from Zcash’s knowledge graph. Always available, no matter where you’re reading.
Why Zcash Needs This Now
The Complexity Gap Is Growing
Halo 2, Orchard, recursive proofs. Each cryptographic advance raises the bar for understanding. Without better knowledge infrastructure, you’re asking developers to climb an increasingly steep mountain.
Developer Time Is Scarce
Every hour spent searching is an hour not building. Every contributor who gives up because they can’t figure something out is a lost opportunity. Every question asked twice in Discord is wasted time. The inefficiency compounds.
Onboarding Is Everything
Going from “interested in privacy tech” to “shipping Zcash applications” is too hard right now. Dead-ends and outdated information kill momentum. Making onboarding seamless directly translates to ecosystem growth.
Please see the full proposal below.
Application Owners
@deusmachinea
@dev-cj
@nitesh32,
@bharatsachya
Organization Name
Bytebell.ai
How did you learn about Zcash Community Grants
ZCash website
Requested Grant Amount (USD)
$30500
Category
Education
Project Lead
Name: Saurav verma
Role: CoFounder and CEO
Background:
I’m Saurav, IIT Delhi ’08, a builder since 2008. I led infrastructure and reliability at Polygon and Biconomy, taking brittle systems to low-latency, high-scale production. I sold my first startup in 2012 and have built six startups that did not work out.
Mad Machines (2013–2015, review summarization with ML). Open sourced and shut down.
Ledgerbus (2015–2016, data sharing between AI agents). Shut down.
DataPod (2016–2017, personal data hosting on local machines). Open sourced and shut down.
ama.fans (2022–2023, personal inbox on blockchain). Open sourced and shut down.
Ping Box (2022–2023, wallet-to-wallet messaging). Open sourced and shut down.
Uniping (2023–2024, ad network on blockchain). Open sourced and shut down.
pyano.network (2024–2025, on-device AI with a coding copilot). Open sourced and pivoted to Bytebell.
Now I’m focused on Bytebell (2025–present).
My learning is simple. Customers do not buy shiny tech. They buy tools that solve problems in time, money, or entertainment
Responsibilities: Architecture, Design and code.
Additional Team Members
- Name: Chaitany Jatoliya
Role: Co-founder and Head of Engineering
Background: 7 years of engineering, Two web3 startups.
Responsibilities: Helps in developing, designing and maintaining the whole Copilot.
- Name: Lovanshu
Role: Developer
Background: Have 1+ years of experience
Responsibilities: Helps in developing frontend code for AI Engine.
- Name: Nitesh
Role: Developer
Background:
Responsibilities: Helps in developing backend code for our Core AI Engine
- Name: Kavish Shah
Role: Developer
Background: Fresh out of college
Responsibilities: Helps in developing the backend code for our Core AI Engine.
Project Summary
Bytebell is a developer copilot that prevents technical information scattering. It ingests code, docs, and conversations, builds a version-aware graph, and returns answers that link directly to the right file, line, and commit. If it cannot prove the answer, it will not respond.
Project Description
Bytebell is built around a simple idea. Every company bleeds time because technical knowledge keeps scattering. Code lives in one place, docs in another, answers in Discord threads, and what people learn in chat never makes it to the docs. Developers spend hours searching, context switching, and repeating what someone already solved. Over time, teams forget how things actually work.
We built Bytebell to stop that. It is a developer copilot that reads code, docs, and conversations and connects them into one live knowledge graph. When someone asks a question, it gives an answer backed by proof—exact file, line, branch, or release. If it cannot prove it, it stays quiet. That small rule changes everything. It means teams can finally trust what they read.
The goal is bigger than just search. We want to turn every team’s scattered information into a shared memory that grows stronger with use. Bytebell fits where developers already work—inside the IDE, Slack, CLI, and browser. It keeps learning how the team builds and documents things, so next time someone asks the same question, the answer is instant and verified.
Over time, Bytebell becomes the single source of truth inside the company. It saves hours, prevents repeated mistakes, and brings clarity where there was noise. That is what we mean when we say we prevent technical information from scattering.
Proposed Problem
It is pretty common for every company to waste a ton of time because technical knowledge just gets scattered everywhere. The code is stashed in one spot, the documentation is tucked away in another, important answers are buried in old Discord threads, and the insights people pick up from casual chats never seem to make it into the official docs. As a result, developers end up spending hours just digging around, constantly switching between tasks, and redoing work that’s already been solved by someone else. Eventually, the whole team starts losing track of how things actually function.
Proposed Solution
We ingest code and documents under your existing permissions, build a version-aware graph (files, sections, commits, releases), and return answers cited to the exact file, line, branch, and release. If we can’t prove it, we don’t answer. This cuts repeat questions, speeds onboarding, keeps docs in sync with code, and gives leaders auditability and clear ROI.
Solution Format
Primary deliverable: production software
- Bytebell cloud app (and private cloud/VPC option): ingestion, search, receipts, analytics.
- MCP server so any assistant/agent can pull governed answers with tools like search_with_receipts, open_at_line, show_diff, doc_sync_pr.
- Client surfaces: MCP, Slack app, CLI, web app.
- Embeddable site widget and chat agent for docs/support deflection.
- Auto-documentation: detect code changes, open PRs to refresh stale pages with cited diffs.
- Admin & analytics: DAU/WAU, queries per user, citation rate, acceptance rate, time to first correct answer, repeat-question deflection, cost/answer.
- Permission inheritance, audit export, retention controls, signed receipts.
Dependencies
As such, we have no dependencies. Whatever is public, we can ingest and provide. If the ZCash team wants us to do a private deployment for only the team, we can also do that, depending upon the grants.
Technical Approach
We use a graph-RAG approach with our own chunkers and metadata generators. We break code, docs, PDFs, issues, and chats into smart chunks, attach rich metadata, and store them for fast retrieval. The graph links files, sections, symbols, commits, and releases so answers stay tied to the right place.
For parsing, we run OCR for PDFs and images, using OCR models and Gemini for image understanding where it helps. We retrieve with sparse and vector search, then apply reranker models to pull the highest-signal spans before we answer.
We run small helper models to classify the user query, enrich it with synonyms and symbols, and route it to the right sources. The final response is produced through our VolAgent agentic framework, which follows a refusal policy and only answers when it can cite the exact sources.
Upstream Merge Opportunities
- Which upstream Zcash repositories you plan to fork/modify
We plan to use almost all zcash repositories and read them to index.
-What changes you plan to make
None
-
Whether these changes could benefit the wider Zcash ecosystem if merged upstream
Not required -
Any coordination needed with upstream maintainers
No -
Timeline considerations for potential upstream merges
Not applicable
Hardware/Software Costs (USD)
5000
Hardware/Software Justification:
- Cloud Infrastructure (AWS): EC2 instances for hosting the Bytebell copilot application, load balancers, and API endpoints ($2,500)
- Database & Storage: Vector database (Pinecone/Weaviate), PostgreSQL for metadata, S3 storage for indexed documents and embeddings ($1,500)
- Development Tools: GitHub Copilot licenses, monitoring tools (Datadog/New Relic), CI/CD infrastructure ($1,000)
Service Costs (USD):
2500
Service Costs Justification:
- AI Model APIs: OpenAI GPT-4/Claude API calls for answer generation, OpenRouter credits for model diversity testing ($1,500)
- OCR & Vision Services: Gemini API for PDF/image understanding, specialized OCR models for technical diagrams ($500)
- Reranker & Embedding Models: Cohere reranker API, embedding model inference costs for semantic search ($500)
- Crawlers
Note: We use both open-source (self-hosted) and closed-source (API) models to optimize for cost vs. performance
Compensation Costs (USD)
23000
Compensation Costs Justification:
Our total team burn cost is around $10,000/month. We will allocate compensation as follows:
-
Implementation & Integration (2 weeks): $5,000
- Ingesting all Zcash resources (GitHub, docs, blogs, research papers)
- Developing additional features (ZK tutorial generation, chat analytics)
- Custom integrations for Zcash ecosystem
-
Ongoing Development & Maintenance (12 months): $8,000
- Continuous feature improvements and bug fixes
- Real-time index updates and system monitoring
- User support and documentation updates
-
IP Development & R&D (pre-grant and ongoing): $10,000
- Proprietary Graph-RAG architecture development
- Custom chunking algorithms and metadata generators
- VolAgent agentic framework and refusal policy systems
- Ongoing research into improved retrieval and reasoning strategies
- These IP assets, developed over 6+ months, are being contributed to the Zcash ecosystem
Total Budget (USD)
30500
Previous Funding
No
Previous Funding Details
No
Other Funding Sources
No
Other Funding Sources Details
No
Implementation Risks
Technical Risks:
-
Real-time Sync Latency (Medium Risk)
GitHub webhook delays or API rate limits could affect 5-minute indexing SLA
Mitigation: Implement queuing system with fallback polling; adjust SLA to p95 metric (5 min) vs absolute -
Citation Accuracy at Scale (Low-Medium Risk)
As corpus grows, maintaining >90% citation rate becomes harder
Mitigation: Continuous evaluation pipeline with human-in-the-loop review for ambiguous cases
-Model Hallucination (Low Risk)
Despite refusal policies, edge cases may produce incorrect answers
Mitigation: 6 months of testing reduced hallucinations to <5%; admin panel flags suspicious responses for review
We will keep on improving this as the goal is to provide the best AI learning and programming resource for Zcash.
Potential Side Effects
No negative effects, huge upsiode for new developers and huge upside for internal Zcash memebers because they wouldnt have to scan through multiple reousces to get the right information.
Success Metrics
This URL will have AI answer all the possible questions that users will have about Zcash Technology.
We also provide an Admin panel where you could see from which country users are asking the questions on the Dopilot widget, Copilot UI or MCP, and if our AI has answered the questions correctly or not, or whether it hallucinated or not. Since we have been working on it for the last six months, hallucinations chances are below minimal, but if there are documentation gaps, the admin panel will also reflect those.
So the success of the project will be 100% when no single question goes unanswered.
Startup Funding (USD)
15500
Startup Funding Justification
We need to index all the Zcash resources to kick off for which we need to deploy the resources and the cloud and start using the AI models available on the open router.
Milestone Details
- Milestone: 1
Amount (USD): 15500
Expected Completion Date: 2025-11-1
User Stories:
- "As a wallet developer, I want ByteBell embedded on Zcash docs so that I can get instant answers without leaving the documentation site."
- "As a DevRel lead, I want an admin panel to manually update and add resources so that ByteBell stays current with the latest Zcash developments."
- "As an exchange integrator, I want to search across extended Zcash resources (ZIPs, forum discussions) so that I get comprehensive guidance."
Deliverables:
- Indexing of additional Zcash resources: repository, Zcash documentation, pdfs, blogs
- Embeddable JavaScript widget for zcash docs with receipt-backed answers
- Admin dashboard for manual resource management (add/update/remove indexed content, refresh specific repos)
- Support for indexing custom Markdown files, PDFs, and external documentation links
Acceptance Criteria:
- "Widget successfully embedded on 4+ Zcash documentation sites with <2s load time"
- "Admin panel allows adding new resources with indexing completed within 5 minutes"
- "Extended corpus includes 50+ additional documents with ≥90% citation rate"
- "Widget handles 10+ concurrent users without degradation"
- Milestone: 2
Amount (USD): 5000
Expected Completion Date: 2025-11-15
User Stories:
- "As a Zcash core developer, I want MCP client integration so that I can query ByteBell directly from my development workflow."
- "As a documentation maintainer, I want automatic documentation updates when code changes so that docs never drift from implementation."
- "As a wallet SDK maintainer, I want auto-generated docs on every release so that integration guides stay accurate."
Deliverables:
- Model Context Protocol (MCP) client for ByteBell integration with Claude, Cursor, and other MCP-compatible tools
- GitHub Actions workflow for automatic documentation generation on push to specified branches (main, develop, release/*)
- Auto-regeneration of API references, code examples, and integration guides when underlying code changes
- Webhook-based triggers for real-time documentation updates
Acceptance Criteria:
- "MCP client successfully integrates with Cursor IDE and responds to queries within 3s"
- "Auto-docs generated within 10 minutes of code push to monitored branches"
- "Generated documentation includes working code examples tested against current codebase"
- "Webhook triggers fire successfully for 95% of qualifying commits"
- Milestone: 3
Amount (USD): 5000
Expected Completion Date: 2025-12-15
User Stories:
- "As a Zcash protocol researcher, I want real-time access to the latest code changes so that my answers reflect bleeding-edge development."
- "As a security auditor, I want to query recent commits and PRs so that I can understand context behind changes."
- "As a core contributor, I want ByteBell to know about open issues and discussions so that I can reference current development priorities."
Deliverables:
- Real-time GitHub ingestion pipeline for Zcash repositories (commits, PRs, issues, discussions)
- Incremental indexing system that updates within 5 minutes of new commits/PRs
- Support for querying specific branches, commit ranges, and PR diffs
- Integration with GitHub API for metadata (author, timestamp, linked issues, review comments)
- Real-time sync for documentation repositories (zips.z.cash, zcash-docs, etc.)
Acceptance Criteria:
- "New commits indexed and queryable within 5 minutes of push"
- "ByteBell accurately answers questions about PRs merged in the last 24 hours"
- "System handles 50+ repository updates per day without queue buildup"
- "Citation links point to specific commit SHAs and PR numbers with ≥95% accuracy"
- Milestone: 4
Amount (USD): 2500
Expected Completion Date: 2026-01-15
User Stories:
- "As a DevRel manager, I want analytics on what developers are asking so that I can identify documentation gaps."
- "As a community lead, I want to see geographic distribution of queries so that I can focus outreach efforts."
- "As a support engineer, I want to track repeat questions so that I can create targeted FAQs."
- "As a product manager, I want to understand user engagement patterns so that I can improve the copilot experience."
Deliverables:
- Comprehensive analytics dashboard with:
- Query volume, topics, and trends over time
- Geographic distribution of users (based on IP/timezone)
- Session duration, queries per session, and return user rate
- Most asked questions, unanswered queries, and topic clusters
- Citation click-through rates and answer acceptance metrics
- Optional user authentication system (OAuth with GitHub, email/password)
- User session tracking with privacy controls (anonymous vs. authenticated analytics)
- Exportable reports (CSV, JSON) for query logs and usage metrics
- Real-time dashboard updates with filters by date range, topic, user cohort
Acceptance Criteria:
- "Analytics dashboard processes and displays data for 10,000+ queries with <3s load time"
- "Top 20 most-asked questions identified automatically with 90% accuracy"
- "User login system supports GitHub OAuth with <5s authentication flow"
- "Geographic analytics show user distribution across ≥15 countries in pilot period"
- "Exportable reports include timestamp, query text, citations shown, user engagement metrics"
- Milestone: 5
Amount (USD): 2500
Expected Completion Date: 2026-11-1
User Stories:
- "As Zcash ecosystem leadership, I want a comprehensive impact report so that we can demonstrate ROI to stakeholders."
- "As a potential adopter, I want to see quantified benefits so that I can justify ByteBell for my project."
- "As the grant committee, I want proof of developer adoption and satisfaction so that we can evaluate success."
Deliverables:
- Complete 12-month impact report including:
- Total queries handled, unique users, query resolution rate
- Average time-to-answer vs. traditional support channels
- Documentation gap analysis (most-asked topics without good docs)
- Developer satisfaction survey results (NPS, CSAT scores)
- Case studies: 2-3 specific examples of developer onboarding acceleration
- Production-ready system with 99.5% uptime over final 3 months
- Sustainability plan: cost projections, scaling guidelines, recommended maintenance schedule
Acceptance Criteria:
- "System successfully handled ≥25,000 total queries with ≥92% citation rate"
- "Average query resolution time <4s (p95 latency)"
- "Documented reduction in support ticket volume by ≥30% for covered topics"
- "Case studies show median first-PR time for new contributors reduced by ≥20%"
- "System maintained 99.5% uptime in final quarter"
- "Complete documentation enables Zcash team to operate ByteBell independently"
Earlier, we submitted a proposal of $20,000, but we redrafted it to account for 1 year of operational costs also and are reapplying with a $30,500 grantA working version is already live at https://zcash.bytebell.ai
We will be in constant touch with the team and also with the community here on the Zcash forum to get feedback. We will improve this product until writing, deploying, and learning Zcash becomes a slice of decentralized private cake.
Please help us with our endeavors.
