Skip to main content



Agentic AI Development- Optimization- Bedrock Claude


Most of the models which you wish to invoke via AWS Bedrock, GPT, Google, etc, charges per token ( words/ characters )

For example - 

A model like Claude is not a computer program that executes instructions. It is a statistical pattern matcher trained on text.

It has no CPU, no memory, no ability to run code. It only predicts the next most likely token based on patterns it saw during training.


But, some AI platforms give the model a sandboxed Python environment to actually execute code. Bedrock does have a Code Interpreter tool in Bedrock Agents — but that's a completely different setup- the overhead of setting it up just to decompress JSON makes zero sense when simply sending clean JSON is trivially easy.


Easy Way To Estimate Your Cost-

Step 1 — Measure your actual prompt size:

  • Take your evidence JSON + instructions
  • Paste into: https://platform.openai.com/tokenizer
  • (OpenAI's tokenizer is close enough for estimation)
  • It will tell you exact token count
Step2 - Cost calculation
  • Cost = (input_tokens / 1000 × 0.003) + (output_tokens / 1000 × 0.015)
Step 3 - Multiply the volume or number of invocations
  • Cost per case × cases per month = monthly Bedrock bill

Reduce Cost - 
  •  Strip null fields from evidence JSON     → fewer tokens
  •  Shorten field names                      → fewer tokens  
  •  Drop fields Claude doesn't need          → fewer tokens
  •  System prompt caching                    → Bedrock charges less for repeated instructions
These reduce the per invocation with token count without any architectural complexity.

Comments

Post a Comment

Popular posts

Unix Server ( Edge Node ) hangs when there are many jobs running on hadoop cluster started from Unix Edge Node.

  When a unix server or an edge node is running lots of jobs (like Spark, Hadoop, or custom batch processes), crashes happen. For example. For example a process might hit a segementation fault, memory issue or ay other runtime issue. By default, if ulimit -c is not 0, the OS will create core dump. Core dump are written to disk and can be very large, sometimes hundreds of MBs or even GBs per process. What we realized was that when multiple processes crash at the same time, the system suddenly tries to write core files to disk. This was leading to DisK I/O spikes. Thus, node was becoming unresponsive. This was also leading CPU spike because OS was handling crash logging. Setting "ulimit -c 0" disables core dumps. This way we lose ability to debug crashes via core dump But, kept production edge nodes stable. On most Linux systems, by default, "core dumps" are written in current working directory of the process that crashes. Linux allows you to change core dump file nam...
Post a Comment
Read more




VPN testing checklist

 To test: Is VPN active? Is it leaking? Does it look residential? Does it alter network fingerprints? Public IP Test (Basic Detection) - curl ipinfo.io DNS Leak Test - nslookup google.com MTU Test (Encapsulation Detection) - ping 8.8.8.8 -f -l 1472 TCP MSS Inspection - Use Wireshark and Look for SYN packets. Traceroute Path Test - tracert 8.8.8.8 WebRTC Leak Test - browserleaks.com/webrtc IPv6 Leak Test - test-ipv6.com Latency / Jitter Test - ping -n 50 8.8.8.8. VPN usually causes: More Jitters and Latency. Reverse DNS (PTR Record) - nslookup <your-public-ip> IP Reputation Check - Like most IP's from VPN Providers like Express VPN, etc. are blacklisted and all. Look for: Proxy/VPN classification, Hosting provider tagging
Post a Comment
Read more




Spark-JDBC connection with Oracle Fails - java.sql.SQLSyntaxErrorException: ORA-00903: invalid table name

  While connecting Spark with Oracle JDBC, one may observe exception like below -  spark.read.format("jdbc"). option("url", "jdbc:oracle:thin:@//oraclehost:1521/servicename"). option("dbtable", "mytable"). option("user", "myuser").option("driver", "oracle.jdbc.driver.OracleDriver") option("password", "mypassword"). load().write.parquet("/data/out") java.sql.SQLSyntaxErrorException: ORA-00903: invalid table name at oracle.jdbc.driver.T4CTTIoer.processError(T4CTTIoer.java:447) at oracle.jdbc.driver.T4CTTIoer.processError(T4CTTIoer.java:396) at oracle.jdbc.driver.T4C8Oall.processError(T4C8Oall.java:951) at oracle.jdbc.driver.T4CTTIfun.receive(T4CTTIfun.java:513) at oracle.jdbc.driver.T4CTTIfun.doRPC(T4CTTIfun.java:227) at oracle.jdbc.driver.T4C8Oall.doOALL(T4C8Oall.java:531) at oracle.jdbc.driver.T4CPreparedStatement.doOall8(T4CPreparedStatement.java:208) ...
Post a Comment
Read more




Spring MongoDB Rest API not returning response in 90 seconds which is leading to client timeout

  We have Spring Boot  Rest API deployed in Kubernetes cluster which integrates with MongoDB to fetch the data.  MongoDB is fed with data by a real time Spark & NiFi job.  Our clients complained that for a request what they send they don't have response within 90 seconds. Consider it like an OMS ( Order ManagEment System).  On further analysis, we found that Spark & NiFi processing is happenning within 10 seconds after consuming response data from Kafka. Thus, initally out thought was that it due to delay from upstream to produce data in to Kafka.  Thankfully, our data had create / request  timestamp, and when response was received, and when response was inserted into MongoDB. Subtracting response insert time from request time seemed to be well within 90 seconds. But, still client did timeout on not seeing a response within 90 seconds. This led to confusion on our side.  But, then we realized it was due to Read Preference . We updated this...
Post a Comment
Read more