Suprovo Mallick (DYNO) DYNOSuprovo

`01 // NEURAL_DOSSIER [IDENTITY_MATRIX]`

# OPERATOR IDENTITY SPECIFICATION
OPERATOR:        Suprovo Mallick
CALLSIGN:        DYNO
ROLE:            Generative AI & ML Systems Engineer
ACADEMICS:       B.Tech in Computer Science @ KIIT (2023 - 2027)
ACCREDITATION:   BS in Data Science (Foundation) @ IIT Madras
COORDINATES:     West Bengal, India [88.3639° E, 22.5726° N]

# COMBAT MANIFESTO
DIRECTIVE: >
  "Notebook prototypes are child's play. I engineer verifiable,
  evidence-grounded, low-latency AI pipelines and multi-agent
  swarms architected to withstand heavy production concurrency."

⚡ `TELEMETRY INTEGRITY GAUGES`

Adaptive RAG & Dynamic Pruning [████████████████] 100% (NLI Filtering)
Multi-Agent State Orchestration [██████████████░░] 92% (LangGraph StateGraphs)
Deterministic Output Grounding [██████████████░░] 90% (Token-Overlap Bypass)
Inference Concurrency & APIs [███████████████░] 95% (FastAPI / 4-bit Quantization)
Open Source Runtime Patches [████████████████] 98% (ComfyUI / LlamaIndex / Haystack)

[RUNTIME_STATUS]
RETRIEVAL_LATENCY   = < 2.0s
HALLUCINATION_RATE  = < 10%
TOKEN_OPTIMIZATION  = ~80% PRUNED
CITATION_PRECISION  = 100% VERIFIED

`02 // ARCHITECTURAL_CORE [EVIDENCE_FIRST_AI]`

"Foundation models are commoditized runtime engines. Real AI engineering begins at query semantics, dynamic pruning, context sanitation, and deterministic output verification."

flowchart TD
    classDef influx fill:#050816,stroke:#00f7ff,stroke-width:2px,color:#00f7ff;
    classDef router fill:#120136,stroke:#7209b7,stroke-width:2px,color:#fff;
    classDef filter fill:#240046,stroke:#ff007f,stroke-width:2px,color:#ff70a6;
    classDef mesh fill:#0a192f,stroke:#00ff99,stroke-width:2px,color:#00ff99;
    classDef gate fill:#1f1147,stroke:#ffd166,stroke-width:2px,color:#ffd166;
    classDef out fill:#003566,stroke:#00e5ff,stroke-width:2px,color:#fff;

    Q[User / Telemetry Stream Influx]:::influx --> R[Semantic Router Node]:::router
    R --> S[Hybrid Dense-Sparse Fetch: ChromaDB + BM25]:::router
    S --> P[NLI Context Pruning Engine: Threshold >= 0.70]:::filter
    P --> M[Multi-Agent Reasoning Graph: LangGraph]:::mesh
    M --> V{Deterministic Gate: Overlap Heuristic >= 0.72?}:::gate
    V -- Bypass Heavy LLM --> O[Async FastAPI REST Endpoint]:::out
    V -- Ambiguous Match --> C[Deterministic Critic Node]:::filter
    C --> O

LAYER	ENGINE SUBSYSTEM	ARCHITECTURAL IMPACT
01	Hybrid Vector-Sparse Indexing	Combined ChromaDB dense embeddings with BM25 sparse keyword rankers for multi-modal recall.
02	Cross-Encoder NLI Pruning	Strips duplicate and contradictory claims with NLI score $\ge 0.70$, slashing token spend by $\sim 80%$.
03	Stateful Multi-Agent Mesh	LangGraph orchestration isolating Router, Extractor, Synthesizer, and Critic nodes with cyclic recovery.
04	Deterministic Verification Gate	Token-overlap heuristic ($\ge 0.72$) eliminates costly validation LLM calls; bounds hallucinations to $< 10%$.
05	Low-Latency Runtime Delivery	Multi-worker async FastAPI backend with containerized Docker deployment delivering $< 2.0\text{s}$ execution.

`03 // FLAGSHIP_SYSTEMS [PRODUCTION_NODES]`

🧬 `NODE 01 // ADAPTIVE EVIDENCE RAG`

Intelligent, Self-Evaluating Context-Pruning Retrieval Engine
Latency: < 2.0s • Token Pruning: ~80% • Hallucination Rate: < 10% • EM Accuracy: > 75%

[ARCHITECTURE FLOW]
Query ──► Dynamic Expansion ──► 5x Parallel Fetch ──► NLI Pruning (>=0.70) ──► Grounded Qwen-4bit

Hallucination Suppression ($< 10%$): Employs strict NLI contradiction filtering and atomic claim extraction prior to synthesis.
Context Sanitation: Cross-encoder NLI independence metric eliminates $\sim 80%$ of redundant tokens, ensuring ultra-dense context windows.
Full-Stack Container: Containerized React.js + FastAPI pipeline powered by a local 4-bit quantized Qwen model for sub-2-second inference.

🌌 `NODE 02 // PLUTO v2`

Multi-Agent Document Intelligence & Evidence Verification Mesh
Concurrency: 8-Worker Pool • Extraction Latency Drop: ~80% • Citation Precision: 100%

[MULTI-AGENT PIPELINE]
Document ──► Router Node ──► Extractor + Retriever ──► Synthesizer ──► Critic Gate ──► Verified Payload

100% Verifiable Quote Attribution: Router $\to$ Extractor $\to$ Synthesizer $\to$ Critic state graph guarantees unbacked claims are scrubbed.
80% Latency Drop: Multi-worker thread pools combined with an internal two-tier LLM response cache eliminate redundant computation.
Heuristic Bypass Gate: Token-overlap heuristic ($\ge 0.72$) bypasses heavy LLM critic verification for obvious evidence matches.

🧠 `NODE 03 // INTENT COMPILER`

Autonomous Architecture, Schema & Technical Specification Engine
Engine: LangGraph StateGraph • Inference: Groq LLaMA-3 LPU • Stack: Streamlit + Python

[AUTONOMOUS COMPILATION]
Product Vision ──► Agent Graph ──► Requirements Spec + Architecture Blueprints + DB Schemas + Pseudocode

Compiles natural-language product ideas into complete, production-grade technical blueprints, database schemas, and executable pseudocode.
Employs stateful LangGraph agents that autonomously critique technical viability and validate dependency graphs before compilation.

SYSTEM NODE	CORE STACK & ARCHITECTURE	KEY PERFORMANCE METRIC	LIVE ENDPOINT
🌾 WHEAT GUARDIAN	Edge CV Crop Pathology (`EfficientNetV2B2 + OpenCV + FastAPI`)	93%+ Accuracy • $\sim 300\text{ms}$ Latency	DEPLOYED APP
🌍 TRANSLATE-V2	Low-Resource NMT Transformer (`PyTorch + NLLB-200 + FastAPI`)	-38% Latency Gain via Batching	HF SPACE
🥗 AAHAR	Dietary RAG Assistant (`LangChain + Gemini + ChromaDB + React`)	Personalized Multi-Turn Memory	WEB APP
💰 AI EXPENSE ADVISOR	Financial Retrieval Engine (`Streamlit + LangChain + Gemini`)	Semantic Spend Classification	GITHUB REPO

`04 // OPEN_SOURCE_OPS [17+_MERGED_PRS]`

PRODUCTION REPOSITORY	IMPACT HIGHLIGHT & PR MISSION	DOMAIN
`Comfy-Org / ComfyUI` (PR #16346 / #16312)	Runtime Reliability: Guarded `prompt_worker` against unhandled execute exceptions, preventing execution thread crashes.	GenAI Core Engine
`open-webui / open-webui` (PR #30047 / #30006)	Routing & Observability: Extended `parse_custom_headers()` to propagate `model_id` alongside user contexts for downstream routing.	LLM Interface
`run-llama / llama_index`	Context Integrity: Implemented header-aware deterministic chunking and post-RAG verification primitives.	Agentic RAG
`deepset-ai / haystack`	Evidence Validation: Created factual consistency and hallucination scoring modules against source documents.	Search Pipelines
`MakazhanAlpamys / Soup`	Distributed Training: Patched DeepSpeed empty-param-group guards to properly forward transformer model arguments.	Model Training
`smoothAPI` & `Surgite`	Backend Resilience: Built async retry predicates, backoff logic, and extensible model provider wrappers.	Cloud / APIs

`05 // TACTICAL_ARMORY [FULL_SPECTRUM_STACK]`

`06 // INDUSTRY_LOG [EXPERIENCE_&_RESEARCH]`

🧪 `ROLE // AI CHATBOT & GENAI BACKEND DEVELOPER`

FED Society | 2024 — PRESENT

Client ──► [FastAPI Gateway] ──► [Vector DB RAG] ──► [Memory Matrix] ──► Response

Modular FastAPI Microservices: Engineered scalable backends integrating persistent vector databases with multi-turn conversation memory.
30% Latency Reduction: Accelerated end-to-end response pipelines through embedding matrix pooling and selective cross-encoder reranking.
Production LLM Endpoints: Architected robust fallback-enabled inference endpoints powering conversational wellness workflows.

📚 `RESEARCH // RAAHI (PIC-2026)`

Real-Time Transport Tracking in Low-Connectivity Cities
Proceedings of the 15th Project Innovation Contest (PIC-2026)
Satellite Event of ICDCIT-2026, KIIT

status: Peer-Reviewed & Published
selection: Top 40 out of 160 submissions globally
domains: [Distributed Systems, Edge Routing, Telemetry Sync]

Formulated a fault-tolerant telemetry synchronization mechanism maintaining positional continuity across packet-loss-heavy cellular networks.

`07 // ALGORITHMIC_CORE [PROBLEM_SOLVING]`

ALGORITHMIC DOMAIN	MASTERY DEPTH	CORE HEURISTICS & PATTERNS
Dynamic Programming & Graphs	`[██████████████░░] 88%`	State Compression, Topological Sort, Shortest Path (Dijkstra), Memoization
Sliding Window & Two Pointers	`[████████████████] 96%`	Dynamic Subarrays, Monotonic Queues, Influx Stream Optimization
Trees, Heaps & Tries	`[██████████████░░] 90%`	LCA via Binary Lifting, Prefix Search Trees, Priority Cache Scheduling
Asymptotic Bounds & Memory	`[█████████████░░░] 85%`	Space-Complexity Guards, Bit Manipulation, CPU Cache Locality

`08 // CREDENTIALS_VAULT [CERTIFICATIONS]`

ISSUING BODY	SPECIALIZATION / CERTIFICATION	VALIDATION
Udemy	AI Engineer Agentic Track: Complete Agent & MCP Architecture	VERIFY CREDENTIAL
IBM / Coursera	IBM RAG and Agentic AI Professional Specialization	VERIFY CREDENTIAL
AWS / Coursera	AWS Generative AI Applications Specialization	VERIFY CREDENTIAL
AWS / Coursera	AWS Fundamentals Specialization	VERIFY CREDENTIAL
Microsoft / LinkedIn	Career Essentials in Generative AI	VERIFY CREDENTIAL
IIT Madras	Foundation Level Certificate, BS in Data Science & Applications	VERIFY CREDENTIAL

`09 // TELEMETRY_STREAM [REPO_METRICS]`

`10 // COMM_LINK [TRANSMISSION_PORT]`

[SYSTEM_TRANSMISSION]
STATUS   = "LISTENING_ON_ALL_PORTS"
CHANNELS = ["Agentic RAG Architectures", "Multi-Agent System Graphs", "Deterministic AI Verification", "High-Throughput ML Backends"]
QUOTE    = "Logic over guesswork. Evidence over hallucinations. Deployed systems over notebooks."

🏆 GitHub Achievements

Badge	Description	Status
🦈 Pull Shark	Opened pull requests that have been merged	Bronze 🥉 ✅
🎯 YOLO	Merged a pull request without a review	✅ Unlocked
⚡ Quickdraw	Closed an issue or PR within 5 minutes	✅ Unlocked
👥 Pair Extraordinaire	Co-authored commits on merged pull requests	Bronze 🥉 ✅
🧠 Galaxy Brain	Answered discussions with accepted answers	⏳ In progress (8 answers posted)

Navigation Menu

Suprovo Mallick (DYNO) DYNOSuprovo

Achievements

Achievements

Highlights

Block or report DYNOSuprovo

`01 // NEURAL_DOSSIER [IDENTITY_MATRIX]`

⚡ `TELEMETRY INTEGRITY GAUGES`

`02 // ARCHITECTURAL_CORE [EVIDENCE_FIRST_AI]`

`03 // FLAGSHIP_SYSTEMS [PRODUCTION_NODES]`

🧬 `NODE 01 // ADAPTIVE EVIDENCE RAG`

🌌 `NODE 02 // PLUTO v2`

🧠 `NODE 03 // INTENT COMPILER`

`04 // OPEN_SOURCE_OPS [17+_MERGED_PRS]`

`05 // TACTICAL_ARMORY [FULL_SPECTRUM_STACK]`

`06 // INDUSTRY_LOG [EXPERIENCE_&_RESEARCH]`

🧪 `ROLE // AI CHATBOT & GENAI BACKEND DEVELOPER`

📚 `RESEARCH // RAAHI (PIC-2026)`

`07 // ALGORITHMIC_CORE [PROBLEM_SOLVING]`

`08 // CREDENTIALS_VAULT [CERTIFICATIONS]`

`09 // TELEMETRY_STREAM [REPO_METRICS]`

`10 // COMM_LINK [TRANSMISSION_PORT]`

🏆 GitHub Achievements

Popular repositories Loading

Uh oh!

Suprovo Mallick (DYNO) DYNOSuprovo

Achievements

Achievements

Highlights

01 // NEURAL_DOSSIER [IDENTITY_MATRIX]

⚡ TELEMETRY INTEGRITY GAUGES

02 // ARCHITECTURAL_CORE [EVIDENCE_FIRST_AI]

03 // FLAGSHIP_SYSTEMS [PRODUCTION_NODES]

🧬 NODE 01 // ADAPTIVE EVIDENCE RAG

🌌 NODE 02 // PLUTO v2

🧠 NODE 03 // INTENT COMPILER

04 // OPEN_SOURCE_OPS [17+_MERGED_PRS]

05 // TACTICAL_ARMORY [FULL_SPECTRUM_STACK]

06 // INDUSTRY_LOG [EXPERIENCE_&_RESEARCH]

🧪 ROLE // AI CHATBOT & GENAI BACKEND DEVELOPER

📚 RESEARCH // RAAHI (PIC-2026)

07 // ALGORITHMIC_CORE [PROBLEM_SOLVING]

08 // CREDENTIALS_VAULT [CERTIFICATIONS]

09 // TELEMETRY_STREAM [REPO_METRICS]

10 // COMM_LINK [TRANSMISSION_PORT]

🏆 GitHub Achievements

Popular repositories Loading

Uh oh!

`01 // NEURAL_DOSSIER [IDENTITY_MATRIX]`

⚡ `TELEMETRY INTEGRITY GAUGES`

`02 // ARCHITECTURAL_CORE [EVIDENCE_FIRST_AI]`

`03 // FLAGSHIP_SYSTEMS [PRODUCTION_NODES]`

🧬 `NODE 01 // ADAPTIVE EVIDENCE RAG`

🌌 `NODE 02 // PLUTO v2`

🧠 `NODE 03 // INTENT COMPILER`

`04 // OPEN_SOURCE_OPS [17+_MERGED_PRS]`

`05 // TACTICAL_ARMORY [FULL_SPECTRUM_STACK]`

`06 // INDUSTRY_LOG [EXPERIENCE_&_RESEARCH]`

🧪 `ROLE // AI CHATBOT & GENAI BACKEND DEVELOPER`

📚 `RESEARCH // RAAHI (PIC-2026)`

`07 // ALGORITHMIC_CORE [PROBLEM_SOLVING]`

`08 // CREDENTIALS_VAULT [CERTIFICATIONS]`

`09 // TELEMETRY_STREAM [REPO_METRICS]`

`10 // COMM_LINK [TRANSMISSION_PORT]`