|
┌─────────────────────────────────────────────────────────────────────────────┐
│ │
│ Name Rishi Sangare │
│ Role Backend & LLM Systems Engineer │
│ Location Mumbai, India │
│ Education MBATech (CS + Finance) @ NMIMS | 2021-2026 │
│ │
│ Focus ▸ Production LLM Systems │
│ ▸ Search Infrastructure │
│ ▸ Cloud Architecture │
│ ▸ API Pipeline Design │
│ │
│ Philosophy "Treat LLMs as unreliable. Add guardrails. Validate." │
│ │
└─────────────────────────────────────────────────────────────────────────────┘
{
"languages": {
"primary": ["Python", "SQL"],
"secondary": ["Java", "JavaScript", "C++"]
},
"backend": {
"frameworks": ["FastAPI", "Flask"],
"patterns": ["REST APIs", "Async Pipelines", "Event-Driven"]
},
"cloud": {
"aws": ["ECS", "ECR", "Lambda", "API Gateway", "CDK", "CloudWatch"],
"containers": ["Docker", "Docker Compose"],
"other": ["Linode", "RunPod"]
},
"search": {
"engines": ["Elasticsearch", "OpenSearch"],
"techniques": ["Hybrid RAG", "BM25", "Vector Search", "Re-ranking"]
},
"ai_ml": {
"apis": ["OpenAI", "Claude", "Gemini"],
"local": ["Fine-tuning", "Transformers", "Mistral-7B"]
},
"devops": {
"ci_cd": ["GitHub Actions", "Blue-Green Deployment"],
"monitoring": ["CloudWatch", "Slack Alerts", "Dozzle"]
}
}
total 5
drwxr-xr-x recruiter-copilot 4.0K production LLM + Search Pipeline
drwxr-xr-x hybrid-rag 3.2K production RAG Architecture
drwxr-xr-x cloud-infra 2.8K production AWS + Linode CI/CD
drwxr-xr-x chrome-extension 2.1K production LinkedIn Data Pipeline
drwxr-xr-x llm-from-scratch 1.5K learning Transformer Implementation
Production LLM + Search Pipeline — Evaluates 40+ candidates per job description
%%{init: {'theme': 'dark', 'themeVariables': { 'primaryColor': '#00d4ff', 'edgeLabelBackground':'#0d1117', 'tertiaryColor': '#161b22', 'lineColor': '#3fb950', 'textColor': '#c9d1d9'}}}%%
flowchart LR
A[📄 Job Description] --> B[🔍 Feature Extraction]
B --> C[⚡ Query Generation]
C --> D[(OpenSearch)]
D --> E[👥 Candidate Pool]
E --> F[🤖 LLM Evaluation]
F --> G[📊 Ranked Results]
style A fill:#1f6feb,stroke:#00d4ff
style D fill:#238636,stroke:#3fb950
style F fill:#8957e5,stroke:#bc8cff
style G fill:#da3633,stroke:#f85149
| Metric | Before | After |
|---|---|---|
| Latency | 2.5 min |
40 sec |
| Tokens/Batch | — | ~80,000 |
| Candidates | — | ~40/request |
Retrieval Augmented Generation — BM25 + Vector dual retrieval
%%{init: {'theme': 'dark', 'themeVariables': { 'primaryColor': '#3fb950', 'tertiaryColor': '#161b22', 'lineColor': '#00d4ff'}}}%%
flowchart LR
Q[🔎 Query] --> T[Transform]
T --> B[BM25<br/>Lexical]
T --> V[Vector<br/>Semantic]
B --> M[🔀 Merge]
V --> M
M --> R[Re-rank]
R --> C[📝 Context]
C --> L[🤖 LLM]
style B fill:#1f6feb,stroke:#00d4ff
style V fill:#8957e5,stroke:#bc8cff
style M fill:#238636,stroke:#3fb950
style L fill:#da3633,stroke:#f85149
✓ Deterministic Retrieval ✓ Bounded Context ✓ Testable ✓ Anti-Hallucination
AWS + Linode Deployment — Blue-Green CI/CD with zero downtime
%%{init: {'theme': 'dark', 'themeVariables': { 'primaryColor': '#f778ba', 'tertiaryColor': '#161b22', 'lineColor': '#00d4ff'}}}%%
flowchart LR
subgraph AWS ["☁️ AWS Stack"]
A1[GitHub] --> A2[Actions]
A2 --> A3[ECR]
A3 --> A4[ECS Fargate]
end
subgraph LIN ["🖥️ Linode Stack"]
L1[GitHub] --> L2[Actions]
L2 --> L3[🔵 Blue]
L2 --> L4[🟢 Green]
end
style A4 fill:#ff7b00,stroke:#ffa657
style L3 fill:#1f6feb,stroke:#00d4ff
style L4 fill:#238636,stroke:#3fb950
| Environment | Stack |
|---|---|
| AWS | ECS + ECR + CDK + API Gateway |
| Linode | Blue-Green + Health Checks + Rollback |
LinkedIn Data Pipeline — Production tool for recruiters
%%{init: {'theme': 'dark', 'themeVariables': { 'primaryColor': '#ffa657', 'tertiaryColor': '#161b22', 'lineColor': '#3fb950'}}}%%
sequenceDiagram
participant E as 🔌 Extension
participant L as 💼 LinkedIn
participant A as ⚡ Lambda
participant D as 🗄️ DynamoDB
E->>L: Extract Profile
E->>A: POST /candidate
A->>D: Upsert Record
A->>E: Structured Response
| Production Stats | |
|---|---|
| Daily API Hits | ~70-80 |
| Active Users | Sachi (Japan) |
| Uptime | 99.9% |
Transformer Implementation — Following Attention Is All You Need
┌─────────────────────────────────────────────────────────────┐
│ TRANSFORMER BLOCK │
│ ┌───────────────────────────────────────────────────────┐ │
│ │ Multi-Head Self-Attention │ │
│ │ Q ──┐ │ │
│ │ K ──┼──► Scaled Dot-Product ──► Concat │ │
│ │ V ──┘ │ │
│ └───────────────────────────────────────────────────────┘ │
│ │ │
│ Add & Norm │
│ │ │
│ ┌───────────────────────────────────────────────────────┐ │
│ │ Feed Forward Network │ │
│ └───────────────────────────────────────────────────────┘ │
│ │ │
│ Add & Norm │
└─────────────────────────────────────────────────────────────┘
Outcome: Understanding why LLM failures are system-level, not model-level.
+ Feb 2025 LD Technologies ─────────────── Backend & LLM Systems Engineer
│ FastAPI • AWS CDK • OpenSearch • LLM Pipelines • Chrome Extension
Jul 2024 Splan Infocom ──────────────── Backend API Developer
│ Invoice Processing • OpenAI API • AWS EC2
Jun 2024 Paragon Dynamics ───────────── Backend Intern
│ Logistics Chatbot • Ledger Reconciliation
Dec 2023 Mitwa.ai ───────────────────── Founding Engineer
│ MERN Stack • LLM Fine-tuning (Mistral-7B) • RunPod
╔═══════════════════════════════════════════════════════════════════════════╗
║ NMIMS Mumbai ║
║ ├── Degree MBATech (Computer Engineering + Finance) ║
║ ├── Duration 2021 - 2026 ║
║ ├── GPA 3.09 / 4.00 ║
║ └── Courses DSA • ML • AI • DBMS • Financial Planning • Tech Mgmt ║
╠═══════════════════════════════════════════════════════════════════════════╣
║ Certifications ║
║ ├── Generative AI with LLMs ──────── DeepLearning.AI & AWS ║
║ └── Advanced Data Science & AI ───── IIT Madras ║
╚═══════════════════════════════════════════════════════════════════════════╝
┌─────────────────────────────────────────────────────────────────────────────┐
│ Some repositories contain architecture documentation only. │
│ Client source code not shared due to IP policies. │
│ System designs and decisions are real + production-tested. │
│ │
│ Happy to discuss tradeoffs and implementation details. │
└─────────────────────────────────────────────────────────────────────────────┘
└──────────────────────────────────────────────────────────────────────────────────────────────┘