AI Infrastructure Engineer · Agent Systems · Founders Office
Saraswat
Building fast, verifiable AI agent infrastructure.
Agent Infra · ZK-Verified Compute · P2P Mesh · Rust · Post-Quantum
About
Hey, I'm Saraswat — 24.
R&D Lead in a founder's office, building AI automation infrastructure that transforms complex, multi-location EdTech operations into efficient, self-updating systems. Designed and deployed the core data and agent backbone for a 15-location network — unified APIs, multi-model LLM and agent pipelines, OCR, and real-time KPI systems — eliminating around 90% of repetitive operational work for 50+ staff in production. In parallel, working on ZK-proof AI agent verification (RISC Zero, Groth16), Ethereum L1 distributed systems, and open-source contributions to Lighthouse, the leading Rust Ethereum consensus client.
Currently at
AITS Group · Founders Office
R&D Lead
Work I did
AITS Group · Founders Office
R&D Lead — Founders Office
- →Built entire technical infrastructure from zero — PostgreSQL warehouse (30+ tables), 15+ API integrations, Power BI dashboards across 15 EdTech locations
- →Deployed multi-agent LLM pipelines: Ollama + Vector DB → 85% manual overhead eliminated, 50+ staff automated
- →OCR + handwriting recognition pipeline, custom HRMS, real-time KPI monitoring across 100+ workforce
Founders Office · R&D
Research & Development Lead — AI Systems
- →Led R&D on AI agent pipelines: designed multi-LLM orchestration, prompt engineering frameworks, and evaluation systems
- →Built internal AI tools: document intelligence, knowledge base search, automated reporting across departments
- →Researched and integrated emerging AI models (Ollama, open-weight LLMs) into production EdTech workflows
- →Drove tech strategy — evaluated tools, built PoCs, shipped production-grade AI automation for 15-location network
Protocol & Agent Infrastructure
Author + Contributor
- →axiom-engine — ZK-proof AI agent engine: RISC Zero zkVM, Groth16 receipts, ML-KEM-768, AWS Nitro TEE
- →Lighthouse (sigp/lighthouse, 3.4k★) — Ethereum consensus client: RocksDB storage layer contribution
- →dvt-focil, protocol-engine-core, p2pflow, Networking-Agent — all public on GitHub
Competitions & Patents
Builder · Competitor · Inventor
- →SIH 2023 Runner-Up · AICTE IDE 2024 National Champion · IIT Hyderabad ITIC 2024 Champion (₹50K)
- →OCF Ocean Tech — ML marine pollution detection, patent filed with Ocean Conservation Foundation
- →YI Innovation Challenge Winner · L&T Ideathon Special Mention · Cognizant National Finalist
Architecture
AI Agent Infrastructure Stack
How fast, verifiable, sovereign agent systems are built — from task ingestion through parallel execution to cryptographic proof of every action.
Trigger / Task Input
HTTP · webhook · cron · CLI · P2P event
Task Router + Planner
Decomposes goal → sub-tasks, selects agents
MCP Server Layer (rmcp 1.7.0)
Tool registration · TLS auth · rate-limit · audit trail
Parallel Agent Pool
N agents × Tokio tasks — concurrent, no blocking
ZK Proof Layer (axiom-engine)
RISC Zero zkVM · Groth16 receipt per agent action
Verified Output + Audit Log
Result · ZK receipt · replay hash · immutable trace
Vector DB
RAG · memory
P2P Mesh
libp2p · gossipsub
eBPF Monitor
kernel-level audit
Post-Quantum
ML-KEM-768 · ML-DSA-65
Fast parallel execution
Every agent runs as an independent Tokio task. No blocking between agents — sub-tasks fan out concurrently, results merge at the aggregator. Throughput scales horizontally.
Every action produces a ZK receipt
axiom-engine wraps every agent computation in RISC Zero zkVM. The output is not just a result — it is a Groth16 proof that this exact program ran on this exact input. Independently verifiable. Tamper-evident.
MCP protocol layer for tool access
Agents communicate with external tools through rmcp (Model Context Protocol). TLS, rate limiting, PII compliance, and an immutable audit trail baked in — not bolted on.
Kernel-level observability
eBPF CO-RE monitors every TCP connection agents open at the kernel level — before the application layer sees it. No agent can make unauthorized network calls silently.
Post-quantum hardened
All key exchange uses ML-KEM-768 (CRYSTALS-Kyber). All signatures use ML-DSA-65 (CRYSTALS-Dilithium). Agent infrastructure built to survive beyond the quantum threshold.
Parallel Agents
N × async
Tokio tasks
Proof latency
~200ms
Groth16 receipt
P2P fanout
gossipsub
libp2p mesh
Audit trail
100%
ZK-verifiable
Key exchange
ML-KEM-768
post-quantum
Observability
kernel
eBPF CO-RE
15
Locations
50+
Staff Automated
10K+
Users
85%
Overhead Cut
40+
KPI Metrics
5
Hackathon Wins
Experience
Work
R&D Engineer Lead
AITS · Affordable Innovation Techno Service Pvt. Ltd.
EdTech · 15 locations · Tier 3/4 markets · Founders Office
40+
KPIs tracked real-time across 15 sites
85%
Org-wide efficiency improvement
100+
Team members managed across locations
10K+
Students impacted via NEP program
System Architect
Coddle Technologies Pvt. Ltd.
Bengaluru · Remote
- →Designed end-to-end system architecture for product development — scalable, modular, production-ready HLD and LLD blueprints
- →Led technical design reviews: data flow diagrams, API contracts, database schema, service boundaries
- →Architected distributed load balancer with full pipeline schema for US client — designed to handle 1M+ user base at scale
- →Defined system design standards and architectural patterns for the engineering team
- →Collaborated with founders on product roadmap, technical feasibility, and build vs. buy decisions
Projects
What I've Built
AI agent infrastructure core — every agent action produces a ZK execution proof. Cryptographically verifiable, deterministically replayable, post-quantum secured.
Groth16 receipts · egg equality saturation · ML-KEM-768 · AWS Nitro attestation · gossipsub P2P broadcast
Fully autonomous agent infrastructure in Rust — MCP server, Claude AI, SQLite pipeline. Discovers prospects, enriches leads, generates outreach. Zero human intervention.
rmcp 1.7.0 · GitHub/YC search · Hunter.io enrichment · Immutable audit trail · Rate limiting · PII compliance
Distributed coordination substrate for multi-agent systems — BFT consensus, Ethereum protocol components, eBPF P2P observer. The trust layer under agent infrastructure.
2-chain HotStuff · Beacon chain · Block builder pipeline · eBPF P2P observer · 50+ tests
Kernel-level agent network monitor — eBPF CO-RE intercepts every TCP connection at the syscall layer. No agent makes unauthorized calls silently.
Kernel-level packet inspection · Works with Geth · Capability-based execution · Visual traffic demo
Consensus fairness infrastructure — DVT + FOCIL (EIP-7805) with IBFT anchoring. Prevents validator censorship at the agent coordination layer.
Anchor IBFT consensus · FOCIL inclusion list enforcement · Validator inclusion fairness
Ethereum consensus client (3.4k★). OSS contributor — RocksDB storage layer, consensus optimizations.
sigp/lighthouse fork · Active contributor · Storage layer work
Computation layer for running applications inside TEEs with threshold-based multi-party provisioning. No single actor controls the environment.
Quorum Key system · 3-key crypto scheme · Deterministic OCI builds · GCP/Azure/Nitro support · qos_core + qos_host crates
ML-based ocean conservation system — patent filed. Detects, monitors, and predicts marine pollution patterns using computer vision and sensor data.
Patent filed · Ocean Conservation Foundation · Pollution detection pipeline · Marine health scoring
Rysen Learning System
Full-stack EdTech platform with role-based access: Admin, Educator, Principal panels — MCQ engine, leaderboard, video delivery, Prisma DB.
Admin · Educator · Principal roles · MCQTimer · VideoEmbed · Leaderboard · API routes · Production deployed
Open Source
OSS Activity
Protocol engineering work — all verifiable on GitHub.
github.com/Saraswat123 · contribution activity
Achievements
Recognition & Impact
Hackathons
AICTE IDE 2024 — National Champion
1st / 200 teams
Innovation Design & Entrepreneurship · National final
IIT Hyderabad ITIC 2024 — Champion
2nd / 650 teams · ₹50,000 Prize
Vishwakarma Award · ITIC IIT Hyderabad · Water & Sanitation Track
YI Innovation Challenge 2024 — Winner
1st / 100 teams
YUVA Utsav 2024 · Yi Bhubaneswar · Young Indians Innovation
L&T EduTech Ideathon 2024
Top 3 / 450 teams
L&T Construction · World Water Day 2024 · Team SPARKKS
Cognizant Hackathon 2024 — National Finalist
Top 10 / 650 teams
Cognizant National Hackathon · Enterprise track
Patent Filed — OCF Ocean Tech
ML-based Marine Pollution Detection
Ocean Conservation Foundation · Filed
Operational Impact
Manual overhead eliminated in 12 months
Reduction in communication prep across 50+ staff
Improvement in work distribution across 15 sites
Data reporting accuracy in KPI matrix
AITS Founders Office
Building, scaling, and automating — integrating agent-driven workflows to eliminate manual overhead and let teams focus on what matters.
AITS Founders Office
Automating the engine — task orchestration, agent pipelines, and cross-location flow integration that turns 85% of manual work into zero.
Skills
Technical Stack
Agent Infrastructure
Fast Execution Runtime
Verified Compute & Crypto
Protocol & Consensus
Data & Infra
Languages
Writing
Articles & Essays
On AI infrastructure, ZK proofs, privacy, and the philosophy of building.
Vibes
When not writing protocols
lofi hip hop · jazz · ambient · late-night coding sessions
midnight protocol
lofi girl · study beats
Contact
Building systems that drive the future.
Open to Founders Office roles, protocol engineering, and agent infrastructure. Reach out.
saraswatdas94@gmail.com
GitHub
github.com/Saraswat123
linkedin.com/in/saraswat-das
X / Twitter
@SaraswatDas13
Medium
@saraswatdas94
Discord
saraswat4225
Saraswat · 2026 · Built in Rust? No, Next.js. Rust next.