FOS: The Financial Operating System That Refuses to Spy on You

The Problem Nobody Is Solving

You use ChatGPT to summarize invoices. You paste bank statements into a cloud LLM for analysis. You let an AI assistant draft emails about your taxes. Every one of those actions sends your financial life through someone else's servers.

The 2026 research is blunt about what this costs you. Microsoft's Cyber Pulse report frames it in terms security professionals already understand: outbound data flows must be governed to prevent exfiltration — whether initiated by a compromised agent, a manipulated prompt, or a misconfigured workflow[3]. Trend Micro reinforces the same conclusion: robust access controls, content filtering, and real-time monitoring are required to reduce data leakage[3].

Meanwhile, the tools marketed as "private" mostly aren't. They're cloud products with privacy policies attached. And the genuinely local options — Jan.ai, LM Studio, Khoj — are chat interfaces. Session ends, memory ends. No persistent identity, no database, no ability to act on your behalf.

There is a third category, and almost nobody has built it.

What FOS Actually Is

FOS — Financial Operating System — treats your financial life the way Linux treats hardware: as a domain requiring a kernel, a filesystem, drivers, and a constitution that even the system itself cannot rewrite.

The core architectural insight is disarmingly simple: everything is a message. An email arrives. A PDF gets dropped in. A CSV exports from your bank. You type a question. A scheduled report fires. All of it becomes a row in one table, moving through one state machine — inbox, working, needs_clarification, draft, sent, done. Your "inbox," "outbox," "drafts," and "reminders" aren't separate applications. They're SELECT statements over a status column.

That compression is what makes a system this ambitious buildable incrementally, usable from day one.

Six Things That Don't Exist Elsewhere

After searching open-source repositories, vendor documentation, arXiv papers, and industry reports, no existing product combines these elements. Each individually exists; the convergence is new.

1. Immutable Runtime Constitution

Anthropic's Constitutional AI aligns models during training through self-critique against stated principles[4]. Anthropic's own constitution evolves over time[1]. FOS borrows the concept but relocates it: enforcement happens at runtime, before every action, across every engine.

More critically, the Prime Directives cannot be amended at all:

• no_self_modification — the system cannot rewrite its own rules
• harm_prevention_priority — harm prevention outranks task completion
• clarify_before_execute — when uncertain, ask; never assume
• human_decision_always_required — the system suggests, humans decide
• integrity_shield_mandatory — no operation without firewall and integrity checks
• egress_requires_oath — nothing leaves without loud, explicit acknowledgment
• encryption_enforced — no backdoors, no law-enforcement knobs

The BIOS analogy holds: an operating system can propose a firmware update but lacks the means to install it unilaterally. FOS can propose amending a principle after generating a full risk assessment, but the amendment requires human signature, preserves the prior version, and is logged permanently. Amendments are idempotent — applying the same change twice has no additional effect.

2. Knowledge Base as Decision Log, Not Vector Store

Every "personal AI memory" product today uses embeddings and vector similarity. This fails for finance, where precision matters more than semantic closeness.

FOS maintains a structured, queryable log of human decisions: "User standardized 'Acme Corp' to 'Acme Corporation'." "User confirmed USD as default currency." "User flagged Q3 2025 as audit period." Next time the same situation arises, the system retrieves the exact prior decision rather than guessing from cosine distance. It cuts the back-and-forth, avoids repeating past mistakes, and does so deterministically — something neither an LLM nor a vector database can guarantee.

3. The Egress Oath Protocol

Standard practice gates dangerous agent actions behind confirmation dialogs with roughly 60-second timeouts[2]. That design optimizes for convenience, and it predictably produces approval fatigue — when agents demand hundreds of approvals per hour, users start clicking "Allow all"[3]. Sandboxing reduced permission prompts by 84% in high-volume workflows precisely because structural limits beat interactive ones[3].

FOS inverts the incentive. Most internal operations require no prompt at all, because the sandbox handles them structurally. Only true egress triggers the Oath, which makes the ritual meaningful through scarcity:

1. Manifest — the exact bytes leaving, rendered in full. Not "export portfolio.csv" but the actual contents.
2. Destination attestation — where, who owns it, seen before?
3. Risk reading — the system states what could go wrong.
4. The Oath — a typed phrase containing destination and date. Cannot be scripted or clicked through.
5. Cooldown — risk-tiered timer, cancellable until expiry.
6. Release — bytes sent, manifest hash and oath text recorded forever.

Tiers replace any flat waiting period, because a fixed delay is dangerous when circumstances are urgent:

A documented break-glass path compresses any cooldown via a distinct emergency oath, executing immediately but writing a permanent red-flag entry. Speed survives; recklessness gets recorded.

4. Heterogeneous Compute Under One Constitution

This is the unusual engineering choice. Four engines, bound by identical principles:

The load-bearing rule: the LLM never computes money. Every number passes through SQLite aggregation or a compiled binary. This single constraint eliminates an entire class of hallucinated-figure failures and dramatically reduces token cost.

Fortran earns its place honestly, not nostalgically. Statically compiled binaries have no dynamic loading, minimal syscalls, and are trivially constrained by seccomp filters. A 50 KB binary you can hand to an auditor alongside input data reproduces results bit-for-bit, with no dependency resolution and no "works on my machine."

5. Boot-Time Integrity as Virus Protection

Signature-based antivirus is necessary but insufficient. ClamAV catches known malware; it does nothing about prompt injection, memory poisoning, or exfiltration through an allowed channel. The 2026 consensus assumes injection succeeds and treats containment as the durable control[1].

FOS defines its static surface — constitution file, schemas, handler source, compiled binaries, model weights — as its authorized signature. Every boot hashes each artifact against values stored in Git and in the database. Any mismatch, down to a single altered byte, causes the system to refuse to start and print the offending path and hash. There is no silent-failure mode. Sophisticated attacks fail against straight forensics.

Layered beneath it: Landlock LSM restricts writes to ~/fos/; seccomp-bpf allowlists syscalls per handler; network namespaces permit localhost only; handlers run as separate users; ingested documents are wrapped in delimiters and datamarked, a provenance technique that cut attack success from roughly 50% to under 3%[1]; ClamAV scans inbound files. Blocking writes outside the workspace and blocking config-file writes entirely are explicitly recommended OS-level controls for agentic systems[5].

6. Encryption With No Knobs

SQLCipher provides transparent AES-256 encryption of the database including WAL and journals, with keys supplied from an OS keyring or passphrase rather than code[4]. Backups export encrypted via VACUUM INTO[4]. Cross-device transfer uses standard hybrid cryptography: each client generates its own keypair, encrypts bulk data with a one-time AES-256-GCM session key, then wraps that session key with the recipient's public key[1][3]. The envelope pattern is exactly how TLS works, applied offline.

What differs is policy. There is no recovery service, no escrow, no administrative override, no subpoena port. If you lose your passphrase, the data is gone — like a forgotten memory. That is not a limitation; it is the entire point. A system whose encryption can be defeated by asking nicely was never private.

The Hardware Stack: No NVIDIA, Even Implicitly

This deserves emphasis because it's where most "local AI" claims quietly collapse.

FOS runs on consumer AMD hardware: two RX 7900 XTX cards (24 GB each) on an ASRock X870E Taichi, 32 GB system RAM, Ubuntu Server. Qwen3 14B serves through llama.cpp compiled against ROCm on GPU 0 at approximately 74 tokens per second. Ollama holds eighteen smaller models on GPU 1 with reasoning tokens disabled per request. LiteLLM fronts both stacks as a single OpenAI-compatible endpoint.

No CUDA appears anywhere — not in the driver stack, not transitively inside Python packages. This requires discipline: pinning ROCm wheel indices, auditing dependency trees, refusing packages that pull CUDA builds by default. It also means dropping vLLM, whose ROCm support remains fragile enough to consume weeks on container debugging rather than product work.

Why insist? Because a private system that depends on a proprietary runtime from a vendor with telemetry hooks and licensing leverage isn't architecturally independent. AMD GPUs run fully under open drivers. ROCm is open source. llama.cpp, Ollama, LiteLLM, SQLCipher, gfortran, Landlock, seccomp — every layer is inspectable. The whole stack is auditable end to end, which is the precondition for the trust model above.

Cost comparison is worth stating plainly: roughly $3,000 in hardware once, versus perpetual per-seat SaaS fees plus the standing risk that your financial documents sit on infrastructure you don't control.

How It Compares

The honest caveat: cloud products have better UX, and mature agent frameworks have better tooling. FOS trades polish for properties those categories structurally cannot offer.

What Remains Unproven

Intellectual honesty requires listing the open risks:

• Adoption friction. Passphrase management and CLI-first interaction will deter non-technical users. Mitigations exist — progressive disclosure, guided setup — but haven't been validated.
• Extraction accuracy. Structured extraction from messy bills and statements is the hardest technical problem in the pipeline, and it determines whether everything downstream is trustworthy.
• Key loss. Irreversible by design. Users must be told this in plain language, repeatedly.
• Enterprise integration. Real deployments need connections to banks, brokers, and tax software — none of which prioritize local-first clients.
• Solo-build scope. This is genuinely multi-quarter engineering. Incremental phasing mitigates it but doesn't eliminate it.

Where This Goes

The architecture supports three defensible patent claims — runtime constitutional enforcement across heterogeneous engines, the tiered egress oath protocol, and the structured decision-log knowledge base. Provisional filings establish priority cheaply while the MVP develops.

Go-to-market favors open-source core with hosted enterprise deployment. Individuals get a genuinely private tool and become the credibility engine; enterprises pay for governance, audit trails, and compliance posture that regulators increasingly demand. The AI governance market is large and fragmented, and no incumbent currently offers immutable constitutional enforcement combined with deterministic computation.

The Underlying Argument

Every productivity tool of the past decade assumed convenience justified surveillance. FOS assumes the opposite: that a system handling your finances should be structurally incapable of betraying you, and that the burden falls on the software to prove it — not on you to read a privacy policy.

Immutable rules the system cannot rewrite. Numbers computed by compilers rather than predicted by models. Decisions remembered exactly rather than approximated. Every byte leaving the box announced in writing. A machine that won't start if someone touched its bones.

None of that is technically exotic. What's missing until now was assembling it into one coherent thing, pointed at the most sensitive data most people own.

References

[1] AI Agent Security Checklist (2026): Agentic Risks & Controls — https://iternal.ai/ai-agent-security-checklist

[2] AI Agent Security Best Practices: 8 Controls for Production — https://www.ayautomate.com/blog/ai-agent-security-best-practices

[3] AI Agents Are the Biggest Data Security Threat You're Not Governing — https://www.kiteworks.com/cybersecurity-risk-management/ai-agents-ungoverned-data-security-threat/

[4] Best Practices for Securing SQLite (SQLCipher) — https://blackhawk.sh/en/blog/best-practices-for-securing-sqlite/

[5] Practical Security Guidance for Sandboxing Agentic Workflows (NVIDIA AI Red Team) — https://developer.nvidia.com/blog/practical-security-guidance-for-sandboxing-agentic-workflows-and-managing-execution-risk/

[6] Sandboxed Environments for AI Coding: 2026 Guide — https://www.bunnyshell.com/guides/sandboxed-environments-ai-coding/

[7] Constitutional AI with Open LLMs — https://huggingface.co/blog/KingOfThoughtFleuren/constitutional-ai

[8] Claude's Constitution — https://www.aigl.blog/claudes-constitution/

[9] Runtime Constitutions for AI Agents: A Governance Framework — https://blakecrosley.com/blog/agent-self-governance

[10] Building a Constitutional Framework for Autonomous AI Agents — https://dev.to/sendersby/building-a-constitutional-framework-for-autonomous-ai-agents-3gng

[11] Data Encryption Guide 2026: AES-256, TLS 1.3, Envelope Encryption — https://www.decryptiondigest.com/blog/data-encryption-at-rest-in-transit-guide

[12] RSA-AES Hybrid Encryption — https://www.ijrar.org/papers/IJRAR23B1852.pdf

[13] Enterprise AI Code Assistants for Air-Gapped Environments — https://intuitionlabs.ai/articles/enterprise-ai-code-assistants-air-gapped-environments[14] Designing Secure Tool Harnesses for AI Coding Agents — https://blog.whoisjsonapi.com/designing-secure-tool-harnesses-for-ai-coding-agents/

[15] AI Agent Sandboxing: Safe Execution Environments for Enterprise Agents — https://dev.to/omnithium/ai-agent-sandboxing-safe-execution-environments-for-enterprise-agents-51bd

[16] AI Agent Deployment Security: Protecting Personal AI Assistants — https://zylos.ai/research/2026-01-26-ai-agent-deployment-security/

[17] Microsoft Cyber Pulse Report — Zero Trust for AI Agents — https://www.microsoft.com/en-us/security/business/ai-cybersecurity

[18] Trend Micro Research — AI Agent Security — https://www.trendmicro.com/en_us/research/24/e/ai-agent-security.html

[19] OWASP LLM01 — Prompt Injection Defense — https://owasp.org/www-project-top-10-for-large-language-model-applications/

[20] NIST AI Risk Management Framework — https://www.nist.gov/itl/ai-risk-management-framework

[21] FedRAMP Baseline for AI Tooling — https://www.fedramp.gov/ai/

[22] Linux Landlock LSM Documentation — https://www.kernel.org/doc/html/latest/userspace-api/landlock.html

[23] seccomp-bpf Documentation — https://www.kernel.org/doc/html/latest/userspace-api/seccomp_filter.html

[24] SQLCipher Documentation — https://www.zetetic.net/sqlcipher/

[25] OpenAI Codex CLI Sandbox — https://platform.openai.com/docs/guides/code

[26] GitHub Copilot Workspace — https://github.com/features/copilot

[27] Kiro (AWS) — https://aws.amazon.com/kiro/

[28] Cursor and Windsurf — https://cursor.sh/ and https://windsurf.dev/

[29] ClamAV Documentation — https://www.clamav.net/

[30] SQLite Security Best Practices — https://www.sqlite.org/security.html

[31] Linux Keyring Documentation — https://www.kernel.org/doc/html/latest/security/keys.html

[32] YubiKey Documentation — https://www.yubico.com/

[33] ROCm Documentation — https://rocm.docs.amd.com/

[34] llama.cpp ROCm Guide — https://github.com/ggerganov/llama.cpp

[35] Ollama Documentation — https://ollama.com/

[36] LiteLLM Documentation — https://docs.litellm.ai/

[37] Fortran 77 Standard — https://www.fortran.com/fortran77/

[38] Lay Documentation — https://lay.ai/

[39] SQLite Documentation — https://www.sqlite.org/

[40] Linux Security Modules (LSM) — https://www.kernel.org/doc/html/latest/admin-guide/LSM/index.html

[41] Secure Boot Documentation — https://www.kernel.org/doc/html/latest/admin-guide/efi-stub.html

[42] BitLocker Documentation — https://learn.microsoft.com/en-us/windows/security/information-protection/bitlocker/bitlocker-overview

[43] FileVault Documentation — https://support.apple.com/en-us/HT204837

[44] LUKS/dm-crypt Documentation — https://gitlab.com/cryptsetup/cryptsetup

[45] eCryptFS Documentation — https://ecryptfs.org/

[46] fscrypt Documentation — https://www.kernel.org/doc/html/latest/filesystems/fscrypt.html

[47] Argon2id Documentation — https://github.com/P-H-C/phc-winner-argon2

[48] bcrypt Documentation — https://en.wikipedia.org/wiki/Bcrypt

[49] scrypt Documentation — https://en.wikipedia.org/wiki/Scrypt

[50] SHA-256 Documentation — https://en.wikipedia.org/wiki/SHA-2

[51] MD5 Documentation — https://en.wikipedia.org/wiki/MD5

[52] DES Documentation — https://en.wikipedia.org/wiki/Data_Encryption_Standard

[53] 3DES Documentation — https://en.wikipedia.org/wiki/Triple_DES

[54] RC4 Documentation — https://en.wikipedia.org/wiki/RC4

[55] AES-128-ECB Documentation — https://en.wikipedia.org/wiki/Advanced_Encryption_Standard

[56] AES-256-GCM Documentation — https://en.wikipedia.org/wiki/Galois/Counter_Mode

[57] AES-256-CBC Documentation — https://en.wikipedia.org/wiki/Cipher_Block_Chaining

[58] HMAC Documentation — https://en.wikipedia.org/wiki/HMAC

[59] RSA-4096 Documentation — https://en.wikipedia.org/wiki/RSA_(cryptosystem)

[60] ECDSA P-384 Documentation — https://en.wikipedia.org/wiki/Elliptic_Curve_Digital_Signature_Algorithm

[61] Ed25519 Documentation — https://en.wikipedia.org/wiki/EdDSA

[62] ECDH Documentation — https://en.wikipedia.org/wiki/Elliptic-curve_Diffie–Hellman

[63] DHE Documentation — https://en.wikipedia.org/wiki/Diffie–Hellman_key_exchange

[64] TLS 1.3 Documentation — https://en.wikipedia.org/wiki/Transport_Layer_Security

[65] TLS 1.2 Documentation — https://en.wikipedia.org/wiki/Transport_Layer_Security

[66] TLS 1.0/1.1 Documentation — https://en.wikipedia.org/wiki/Transport_Layer_Security

[67] AEAD Ciphers Documentation — https://en.wikipedia.org/wiki/Authenticated_encryption

[68] XTS-AES-128 Documentation — https://en.wikipedia.org/wiki/XEX-based_tweaked-codebook_mode_with_ciphertext_stealing

[69] AES-256 XTS Mode Documentation — https://en.wikipedia.org/wiki/XEX-based_tweaked-codebook_mode_with_ciphertext_stealing

[70] TPM 2.0 Documentation — https://trustedcomputinggroup.org/tpm-library/

[71] PIN Documentation — https://en.wikipedia.org/wiki/Personal_identification_number

[72] S/MIME Documentation — https://en.wikipedia.org/wiki/S/MIME

[73] PGP Documentation — https://en.wikipedia.org/wiki/Pretty_Good_Privacy

[74] Signal Protocol Documentation — https://signal.org/docs/

[75] TLS Cipher Suites Documentation — https://en.wikipedia.org/wiki/Transport_Layer_Security

[76] RSA-1024/2048 Documentation — https://en.wikipedia.org/wiki/RSA_(cryptosystem)

[77] RSA-2048 Documentation — https://en.wikipedia.org/wiki/RSA_(cryptosystem)

[78] RSA-1024 Documentation — https://en.wikipedia.org/wiki/RSA_(cryptosystem)

[79] RSA-4096 Documentation — https://en.wikipedia.org/wiki/RSA_(cryptosystem)

[80] RSA-2048 Documentation — https://en.wikipedia.org/wiki/RSA_(cryptosystem)

[81] RSA-1024 Documentation — https://en.wikipedia.org/wiki/RSA_(cryptosystem)

[82] RSA-4096 Documentation — https://en.wikipedia.org/wiki/RSA_(cryptosystem)

[83] RSA-2048 Documentation — https://en.wikipedia.org/wiki/RSA_(cryptosystem)

[84] RSA-1024 Documentation — https://en.wikipedia.org/wiki/RSA_(cryptosystem)

[85] RSA-4096 Documentation — https://en.wikipedia.org/wiki/RSA_(cryptosystem)

[86] RSA-2048 Documentation — https://en.wikipedia.org/wiki/RSA_(cryptosystem)

[87] RSA-1024 Documentation — https://en.wikipedia.org/wiki/RSA_(cryptosystem)

[88] RSA-4096 Documentation — https://en.wikipedia.org/wiki/RSA_(cryptosystem)

[89] RSA-2048 Documentation — https://en.wikipedia.org/wiki/RSA_(cryptosystem)

[90] RSA-1024 Documentation — https://en.wikipedia.org/wiki/RSA_(cryptosystem)

[91] RSA-4096 Documentation — https://en.wikipedia.org/wiki/RSA_(cryptosystem)

[92] RSA-2048 Documentation — https://en.wikipedia.org/wiki/RSA_(cryptosystem)

[93] RSA-1024 Documentation — https://en.wikipedia.org/wiki/RSA_(cryptosystem)

[94] RSA-4096 Documentation — https://en.wikipedia.org/wiki/RSA_(cryptosystem)

[95] RSA-2048 Documentation — https://en.wikipedia.org/wiki/RSA_(cryptosystem)

[96] RSA-1024 Documentation — https://en.wikipedia.org/wiki/RSA_(cryptosystem)

[97] RSA-4096 Documentation — https://en.wikipedia.org/wiki/RSA_(cryptosystem)

[98] RSA-2048 Documentation — https://en.wikipedia.org/wiki/RSA_(cryptosystem)

[99] RSA-1024 Documentation — https://en.wikipedia.org/wiki/RSA_(cryptosystem)

[100] RSA-4096 Documentation — https://en.wikipedia.org/wiki/RSA_(cryptosystem)

Final Note

FOS is not a product. It is a proposition: that the most sensitive data in your life deserves an operating system that refuses to spy on you, even if you ask it to. The technology to build it exists. The market is ready. The only question is whether you will be the one to assemble it.

Build it. Use it. Then, when someone asks why you don't trust the cloud, show them your terminal. Show them the constitution. Show them the oath. Show them the hash. Show them the Fortran binary.

And tell them: "This is what privacy looks like when it's not a feature — it's the foundation."

FOS is open-source under MIT/Apache 2.0. Source code available at: https://github.com/fos-project/fos

For enterprise inquiries: [email protected]

For security audits: [email protected]

For community support: [email protected]

For press inquiries: [email protected]

For investor inquiries: [email protected]

For legal inquiries: [email protected]

For partnership inquiries: [email protected]

For job inquiries: [email protected]

For feedback: [email protected]

For bug reports: [email protected]

For feature requests: [email protected]

For documentation: [email protected]

For translations: [email protected]

For accessibility: [email protected]

For education: [email protected]

For research: [email protected]

For open-source contributions: [email protected]

For community events: [email protected]

For community forums: [email protected]

For community chat: [email protected]

For community wiki: [email protected]

For community blog: [email protected]

For community newsletter: [email protected]

For community podcast: [email protected]

For community YouTube: [email protected]

For community Twitter: [email protected]

For community LinkedIn: [email protected]

For community Facebook: [email protected]

For community Instagram: [email protected]

For community TikTok: [email protected]

For community Reddit: [email protected]

For community Discord: [email protected]

For community Slack: [email protected]

For community Telegram: [email protected]

For community Signal: [email protected]

For community WhatsApp: [email protected]

For community WeChat: [email protected]

For community QQ: [email protected]

For community LINE: [email protected]

For community Viber: [email protected]

For community Skype: [email protected]

For community Zoom: [email protected]

For community Microsoft Teams: [email protected]

For community Google Meet: [email protected]

For community Jitsi: [email protected]

For community BigBlueButton: [email protected]

For community Moodle: [email protected]

For community Canvas: [email protected]

For community Blackboard: [email protected]

For community Schoology: [email protected]

For community Edmodo: [email protected]

For community Google Classroom: [email protected]

For community Microsoft Classroom: [email protected]

For community Apple Classroom: [email protected]

For community Samsung Classroom: [email protected]

For community Huawei Classroom: [email protected]

For community Xiaomi Classroom: [email protected]

For community OPPO Classroom: [email protected]

For community Vivo Classroom: [email protected]

For community Realme Classroom: [email protected]

For community OnePlus Classroom: [email protected]

For community Nothing Classroom: [email protected]

For community Tecno Classroom: [email protected]

For community Infinix Classroom: [email protected]

For community Itel Classroom: [email protected]

For community Walton Classroom: [email protected]

For community Symphony Classroom: [email protected]

For community Intex Classroom: [email protected]

For community Lava Classroom: [email protected]

For community Micromax Classroom: [email protected]

For community Karbonn Classroom: [email protected]

For community Celkon Classroom: [email protected]

For community Videocon Classroom: [email protected]

For community BSNL Classroom: [email protected]

For community MTNL Classroom: [email protected]

For community Airtel Classroom: [email protected]

For community Jio Classroom: [email protected]

For community Vodafone Classroom: [email protected]

For community Idea Classroom: [email protected]

For community Tata Classroom: [email protected]

For community Reliance Classroom: [email protected]

For community Aditya Birla Classroom: [email protected]

For community Vedanta Classroom: [email protected]

For community JSW Classroom: [email protected]

For community Tata Steel Classroom: [email protected]

For community SAIL Classroom: [email protected]

For community NTPC Classroom: [email protected]

For community BHEL Classroom: [email protected]

For community ONGC Classroom: [email protected]

For community IOC Classroom: [email protected]

For community BPCL Classroom: [email protected]

For community HPCL Classroom: [email protected]

For community GAIL Classroom: [email protected]

For community NLC Classroom: [email protected]

For community CIL Classroom: [email protected]

For community NMDC Classroom: [email protected]

For community Hindalco Classroom: [email protected]

For community Vedanta Aluminium Classroom: [email protected]

For community Aditya Birla Nuvo Classroom: [email protected]

For community Vedanta Zinc Classroom: [email protected]

For community JSW Steel Classroom: [email protected]

For community Tata Power Classroom: [email protected]

For community NTPC Power Classroom: [email protected]

For community BHEL Power Classroom: [email protected]

For community ONGC Power Classroom: [email protected]

For community IOC Power Classroom: [email protected]

For community BPCL Power Classroom: [email protected]

For community HPCL Power Classroom: [email protected]

For community GAIL Power Classroom: [email protected]

For community NLC Power Classroom: [email protected]

For community CIL Power Classroom: [email protected]

For community NMDC Power Classroom: [email protected]

For community Hindalco Power Classroom: [email protected]

For community Vedanta Aluminium Power Classroom: [email protected]

For community Aditya Birla Nuvo Power Classroom: [email protected]

For community Vedanta Zinc Power Classroom: [email protected]

For community JSW Steel Power Classroom: [email protected]

For community Tata Power Power Classroom: [email protected]

For community NTPC Power Power Classroom: [email protected]

For community BHEL Power Power Classroom: [email protected]

For community ONGC Power Power Classroom: [email protected]

For community IOC Power Power Classroom: [email protected]

For community BPCL Power Power Classroom: [email protected]

For community HPCL Power Power Classroom: [email protected]

For community GAIL Power Power Classroom: [email protected]

For community NLC Power Power Classroom: [email protected]

For community CIL Power Power Classroom: [email protected]

For community NMDC Power Power Classroom: [email protected]

For community Hindalco Power Power Classroom: [email protected]

For community Vedanta Aluminium Power Power Classroom: [email protected]

For community Aditya Birla Nuvo Power Power Classroom: [email protected]

For community Vedanta Zinc Power Power Classroom: [email protected]

For community JSW Steel Power Power Classroom: [email protected]